Plain Virtualization

Storage Efficiency • WAFL Internals

2026-09-09T09:00:00.000+08:00

Storage Efficiency • WAFL Internals

Dedupe, Compress, Compact: How WAFL Squeezes Every Byte Out of Capacity Flash

Flash is no longer the exotic performance tier — it’s becoming the default. Which makes the question of how efficiently you use every terabyte of it more important than ever.

🕑 8 min read

A personal note: I’m a presales manager at NetApp, but this blog is my own space for sharing what I find genuinely interesting in storage engineering. The views below are mine alone, not an official NetApp position, and not a sales pitch — just one practitioner nerding out on a filesystem.

WAFL Deduplication Compression Compaction TSSE AFF/ASA C-Series QLC Flash

Every gigabyte of flash you buy today costs more than the equivalent gigabyte of spinning disk, and that gap doesn’t look like it’s closing anytime soon. So the industry’s response has been predictable: build flash platforms that are cheaper per raw terabyte — QLC-based, high-density, "capacity" flash — and then lean hard on software to make every one of those terabytes count. That second half is where things get interesting, and where I think WAFL quietly does some of its best work without anyone talking about it much.

NetApp’s AFF C-Series and ASA C-Series are built exactly for this moment — QLC capacity flash that competes on cost with hybrid arrays while still being all-flash. But the hardware is only half the story. The other half is a set of storage efficiency mechanisms baked directly into WAFL (Write Anywhere File Layout) that have been maturing since ONTAP 8.3, and that I think deserve a closer look than the usual "we do dedupe and compression" bullet point.

A Three-Stage Pipeline, Not Three Separate Features

The thing people often miss is that dedup, compression, and compaction on ONTAP aren’t three independent knobs you flip and hope they add up. They run as an ordered, inline pipeline, each stage cleaning up what the previous one left behind, before a single block ever touches the drive. On an AFF or ASA system the sequence is:

Notice the order: dedup first, then compression, then compaction. That sequencing matters. Deduplication works on whole 4KB blocks, so it needs to run before anything gets shrunk or repacked — otherwise you’d be comparing hashes of compressed fragments instead of the original data, and you’d miss matches. Compaction runs last precisely because its job is to clean up after the other two: it looks at whatever is left over — compressed remnants, small files, oddly-sized I/O — and tries to fit several of those into a single physical block instead of wasting the rest of it.

The 4KB Problem Compaction Was Built to Solve

WAFL has always written in 4KB blocks. That's a deliberate, sensible design choice — it lines up with how flash and modern drives organize physical space. But it creates an obvious inefficiency: if compression shrinks an 8KB write down to 3KB, or if you're writing a pile of small files that are each only 1–2KB, you still end up consuming a full 4KB block for each of them. The leftover space in that block is simply wasted — padding that customers are paying full flash prices for.

Inline data compaction, introduced with ONTAP 9, exists specifically to close that gap. It takes multiple sub-4KB chunks — whether they're small uncompressed files or the leftovers of compression — and packs several of them into a single physical 4KB block before it's ever written to disk. Three separate 1KB writes that would have consumed three full blocks can now share one. It's a genuinely elegant fix for a problem that's specific to fixed-block architectures like WAFL, and it's additive to whatever dedup and compression already saved you.

“Compaction doesn’t compete with dedup or compression — it cleans up after them, turning the space they couldn’t reclaim into usable capacity.”

Efficiency That Runs When the CPU Isn’t Busy

Here's the part I find most underrated: not all of this efficiency work happens inline in the write path. ONTAP also runs storage efficiency as a background process that opportunistically uses spare CPU cycles — the cycles that would otherwise sit idle between bursts of host I/O. This is where Temperature-Sensitive Storage Efficiency (TSSE) comes in, introduced in ONTAP 9.8 and enabled by default on thinly provisioned AFF volumes, and enabled by default across the board on the C-Series.

TSSE works on a simple but clever idea: not all data deserves the same compression effort. Hot, frequently overwritten blocks get compressed in small 8KB groups — fast to process, low latency impact, decent savings. But data that cools off — typically after about two weeks of inactivity — gets picked up by a background scan and recompressed into larger 32KB groups using a stronger algorithm. Bigger groups compress better because there's more redundancy to exploit across a larger window of data, but decompressing them is heavier, which is exactly why you don't want to do it inline on every write. Applying that heavier compression only to cold data, and only when the system has cycles to spare, is a genuinely well-thought-out trade-off between performance and density.

It's worth noting ONTAP's newest AFF platforms (A70, A90, A1K) take a different path — they use a dedicated hardware offload processor to run 32K compression continuously rather than relying on temperature and idle CPU cycles. But for the C-Series capacity flash line specifically, TSSE running on spare CPU is the default mechanism, and it's a big part of why these QLC-based systems can compete on effective cost per terabyte against far less sophisticated capacity architectures.

Why This Is the Whole Point of Capacity Flash

QLC flash is cheaper per raw gigabyte than TLC, but it's also where the "flash is expensive" argument actually gets settled or lost. If a QLC-based array only stored data at its raw density, it wouldn't be competitive against high-capacity hybrid or spinning-disk systems on cost — you'd be paying a flash premium without reaping a flash-level density advantage. The entire economic case for AFF C-Series and ASA C-Series rests on software making the effective capacity meaningfully larger than the raw capacity, consistently, without the customer having to think about it.

That's what this three-stage pipeline plus TSSE is really doing under the hood: dedup removes redundant data, compression shrinks what's left, compaction eliminates the wasted space around fixed 4KB blocks, and TSSE keeps squeezing cold data further over time using capacity the system would otherwise leave idle. None of it requires the customer to run manual jobs, tune schedules, or trade away performance to get it — it's inline and default-on for the platforms built around it.

From a practitioner's seat, this is the detail I try to get customers to look past the headline $/GB number for. Two flash arrays can have identical raw capacity and wildly different effective capacity, and the difference isn't marketing — it's whether the filesystem underneath was actually designed for this problem from the ground up, or bolted efficiency features on afterward.

Curious how this plays out in your own environment? If you're running AFF or ASA today, volume show-footprint is worth a look — it'll show you exactly how much of your footprint reduction is coming from compaction versus compression versus TSSE. I'd genuinely like to hear what ratios people are seeing in the field — drop a comment or find me on LinkedIn.

Disclaimer: This article reflects my personal views and technical understanding as an individual in the storage industry. It does not represent an official position of NetApp or any employer, and it isn't sponsored or reviewed content. Technical details are drawn from publicly available NetApp documentation and community resources; please verify against official docs for your specific ONTAP version before making architecture or purchasing decisions.

References: NetApp ONTAP documentation — Storage efficiency overview, Temperature-sensitive storage efficiency, Deduplication/compression/compaction concepts; NetApp Community — "A Look Inside NetApp Inline Data Compaction."

The Quantum Threat Is Real. Is Your Data Already Protected?

2026-07-28T09:00:00.000+08:00

The Quantum Threat Is Real. Is Your Data Already Protected?

While the industry races to catch up on quantum readiness, NetApp customers have been protected since day one — at the layer that matters most: where your data lives.

⏱ 7 min read

In a world of evolving cyberattacks, organisations invest millions across layers of infrastructure — networks, compute, identity, endpoints — yet a critical question too often goes unasked: when every other layer fails, where do you land? The answer is always the same. Your data. And what is protecting it?

The cybersecurity conversation today is dominated by urgency around quantum computing, ransomware, and zero-trust architectures. Much of the industry is in a reactive mode — bolting on protection after the fact, issuing roadmaps, and promising readiness "soon." Against this backdrop, NetApp stands apart: not because it is catching up, but because it was never behind.

This article makes the case that regardless of where your broader infrastructure security posture stands today, you already have one immovable, validated assurance — if your data runs on NetApp ONTAP, your last line of defence is already the most secure storage on the planet.

The Reality of Infrastructure-Wide Security

No organisation achieves a perfectly hardened security posture overnight. Digital transformation, hybrid cloud adoption, and the sheer complexity of modern IT environments mean security is always a work in progress. Compute platforms add new capabilities; network vendors release patches; identity providers issue updates. That is the nature of defence in depth.

But there is an important difference between components that are working toward security maturity and components that are already certified to the highest global standards. The storage layer — the ultimate destination of all data — should never be the weakest link. With NetApp, it is not.

"Your data is the primary target of ransomware attacks, so you need a storage solution you can trust as your last line of defense. NetApp can safely store and protect your data and help you recover it if an attack occurs." — NetApp Autonomous Ransomware Protection page

Think of it this way: even while other parts of your technology stack are undergoing security upgrades, patching cycles, or standards alignment — your storage layer, if built on NetApp, is already operating at a level that can hold classified national security data. That is not a marketing claim. It is a validated, independently verified fact.

The Quantum Wake-Up Call — and Who Was Already Ready

Post-quantum cryptography (PQC) has recently become a board-level conversation. As quantum computing matures, conventional encryption algorithms — RSA, ECC — face a fundamental vulnerability. The threat is not hypothetical. The so-called "harvest now, decrypt later" strategy is already being used by sophisticated adversaries: encrypted data is stolen today, stored, and held until quantum computing power is sufficient to crack it. The clock is ticking on data you are generating right now.^[14]

Across the infrastructure ecosystem, vendors are publishing roadmaps and blog posts about their plans to become quantum-ready. They are working through architecture redesigns, testing algorithm compatibility, and beginning the long journey toward NIST alignment. This is commendable — but for customers, it means a window of exposure exists today.^[16]

NetApp moved decisively ahead of this curve. In April 2025, NetApp announced that its storage portfolio is compliant with NIST-standardised PQC algorithms, protecting both data at rest and data in transit with quantum-resistant encryption — already embedded in the storage platform, not added on top.^[11]

⚛

PQC: Data at Rest

NIST-approved algorithms embedded in ONTAP for file and block workloads — quantum-safe from the storage OS up.

netapp.com/pqc →

🔒

PQC: Data in Transit

Quantum-safe encryption for data moving across networks — protecting against harvest-now-decrypt-later attacks.

PQC announcement →

📐

NIST Alignment

Full alignment with NIST PQC standards finalised in 2024 — the globally recognised benchmark for quantum-safe encryption.

NetApp blog →

The message for organisations is clear: while other parts of your infrastructure are working toward quantum readiness, your data storage — if on NetApp — is already there. You do not need to wait for every layer to catch up before your most sensitive data is protected.

A Security Pedigree Built Over Two Decades

NetApp's quantum readiness does not exist in isolation. It is the latest expression of a deeply embedded security culture — one that stretches back twenty years and is validated by the most rigorous government and international standards bodies on earth.

Common Criteria: Two Decades of Independent Validation

In 2005, NetApp became the first storage vendor in the world to achieve Common Criteria certification for its core operating system, Data ONTAP.^[3] Common Criteria (ISO/IEC 15408) is the global standard for evaluating IT security products, recognised by 31 countries.^[4] Unlike self-assessed compliance, Common Criteria certifications are awarded by accredited, independent third-party laboratories after rigorous testing. NetApp has continuously maintained and renewed this certification across ONTAP versions, including ONTAP 9 and StorageGRID.^[1] No other storage vendor has matched this two-decade commitment to independently verified security.

NSA CSfC: Trusted to Store Top Secret Data

In 2021, NetApp ONTAP became the first enterprise storage and data management platform to achieve Commercial Solutions for Classified (CSfC) validation from the U.S. National Security Agency.^[17] The CSfC programme is the NSA's commercial cybersecurity strategy for protecting the nation's most sensitive information. CSfC-validated products must implement two independent layers of encryption and pass the most rigorous security requirements for classified National Security Systems data.

The result: NetApp ONTAP is validated and trusted to natively host secret and top-secret data.^[22] If the U.S. intelligence community trusts it with its most classified information, your enterprise and government data is in safe hands.

"This unique achievement is an example of innovation in commercial technology addressing critical national security issues."

Admiral Michael S. Rogers USN (Ret.), former Director, NSA & Chief, Central Security Service

CISA Secure by Design Pledge

In May 2024, NetApp was among the first wave of technology companies to sign the CISA Secure by Design Pledge at RSA Conference — a voluntary commitment to embed security as a foundational design principle across products, not as an add-on.^[35] The pledge, co-developed with the NSA and cybersecurity agencies from the UK, Australia, Canada, and other allied nations, signals the highest level of organisational commitment to proactive security design.

🌎

Common Criteria

First storage vendor certified in 2005. Continuously maintained for 20+ years. Recognised by 31 countries under ISO/IEC 15408.

View certifications →

🔐

NSA CSfC Validation

First enterprise storage platform validated by the NSA to store classified top-secret data. Dual-layer hardware + software encryption.

CSfC details →

📋

CISA Secure by Design

One of the first 68 companies globally to sign the CISA Secure by Design Pledge, co-endorsed by NSA and allied cyber agencies.

Pledge list →

🛡

NIST CSF Alignment

ONTAP's ransomware defence and data protection capabilities map fully to the NIST Cybersecurity Framework: Identify, Protect, Detect, Respond, Recover.

NIST framework blog →

Ransomware Protection: Built In, Not Bolted On

One of the starkest differentiators in the storage security landscape is how ransomware protection is delivered. The predominant approach across the industry is to procure separate security software, integrate it with storage via APIs or agents, and hope the seams do not create blind spots. NetApp took a fundamentally different path.

NetApp Autonomous Ransomware Protection (ARP) is the industry's first — and still the only — AI-powered ransomware detection capability built natively into enterprise primary storage.^[34] Introduced in ONTAP 9.10, ARP uses on-box machine learning to analyse volume workload activity and data entropy in real time. When anomalous patterns consistent with ransomware are detected, it automatically takes a snapshot and alerts administrators — without relying on external agents, signature databases, or bolt-on tools.^[29]

This is not just a design claim — it is independently validated. SE Labs, one of the most respected independent security testing organisations in the world, rigorously tested NetApp ARP/AI against hundreds of known and previously unseen ransomware variants under real-world conditions. The results: 99% detection accuracy with 100% precision — zero false positives. SE Labs awarded NetApp ARP/AI its highest AAA rating, making it the first and only storage vendor to achieve this distinction for on-box ransomware detection.^[36]

Industry norm

Separate security software integrated post-deployment

Agent-based detection with coverage gaps

Reactive — detects after significant encryption

Bolt-on recovery tools, complex orchestration

PQC on the roadmap — not yet delivered

NetApp ONTAP

ARP built natively into the storage OS

On-box ML — no external agents required

Detects after minimal encrypted files, auto-snapshots

Single-pane recovery with 1-click orchestration

PQC compliant — data at rest & in transit, now

🏆

SE Labs AAA Rating

Independent testing against hundreds of ransomware variants: 99% detection accuracy with 100% precision — zero false positives. Highest AAA rating ever awarded to a storage vendor.

SE Labs report →

🥇

SE Labs 2025 Award

Winner of the 2025 SE Labs Award for Enterprise Data Protection — the only storage vendor recognised for AI-powered on-box ransomware detection excellence.

ARP/AI details →

With ARP/AI, NetApp deploys adaptive AI/ML models trained on over a million files, delivering 99% precision and recall, capable of detecting even the newest ransomware variants from the moment they touch the storage layer. In 2025, SE Labs further recognised this by awarding NetApp the SE Labs Award for Enterprise Data Protection — the only storage vendor to win this distinction.^[37]

This is not an incremental improvement. It is a fundamentally different philosophy: security as an intrinsic property of your data infrastructure, not an external system trying to monitor it from the outside. That philosophy is why NetApp was the first storage vendor to offer a financial ransomware recovery guarantee — backing its capabilities with contractual commitment.^[34]

And all of this maps precisely to the NIST Cybersecurity Framework — Identify, Protect, Detect, Respond, Recover — providing organisations with a structured, auditable, and defensible approach to cyber resilience at the storage layer.^[26]

The "Last Line of Defence" Principle in Practice

Security architects speak often about defence in depth — layering controls so that no single failure creates a catastrophic breach. This is sound strategy. But depth requires that each layer is genuinely strong, not just present. When it comes to the storage layer, depth without strength is an illusion.

Consider a practical scenario: an organisation has deployed a modern infrastructure stack — virtualisation, containers, cloud management, identity, network detection. It is in the process of upgrading these components to meet new cryptographic standards. The migration will take months. During that window, is the organisation exposed?

If storage runs on NetApp: no. The data itself — the ultimate target of every attack — is already protected by:

NIST-compliant PQC encryption

NSA-validated dual-layer encryption

AI ransomware detection in storage OS

Common Criteria certified 20+ years

CISA Secure by Design committed

No other storage vendor can make this claim in full. Some have certifications. Some have bolt-on ransomware tools. Some have published PQC roadmaps. NetApp has the credentials, the built-in capabilities, and the validated track record — simultaneously.

This means that even as organisations complete their broader security transformations, they are not gambling with their most critical asset — the data — during the transition. NetApp holds the line.

The Question You Need to Ask Today

Every organisation faces a version of this question: are we going to wait until our entire infrastructure security posture is perfectly aligned before we consider ourselves protected? The answer, in practice, is always no — because perfect posture is never fully achieved. Security is a continuous journey.

The more practical question is: among all the components in your environment, which ones are already operating at the highest proven standard? And: given that your data is the ultimate prize for any attacker, is the layer where it actually lives — your storage — the most secure it can possibly be?

The most battle-hardened, independently verified, government-trusted, quantum-ready, and intrinsically ransomware-protected storage in the world is already available to your organisation. Not as a future promise. Not on a roadmap. Now.

Are you still waiting for all your security layers to be ready before considering yourself protected? Or would you rather know that, no matter what happens across the rest of your stack, the place where your data ultimately lives is the most secured storage on the planet — since day one? Your last line of defence is already secured.

References & Sources

NetApp Community — NetApp StorageGRID earns Common Criteria security certification: community.netapp.com
NetApp Press Release — NetApp Data ONTAP 8.0 Earns Common Criteria Certification: netapp.com
NetApp Trust Centre — Common Criteria Certification & Compliance: netapp.com/common-criteria
NetApp Product Updates — Post-Quantum Cryptography Compliance Announcement: netapp.com/pqc-announcement
NetApp Blog — Why you need post-quantum cryptography at the storage layer: netapp.com/blog
NetApp — Post-Quantum Cryptography product page: netapp.com/pqc
VMware Blog — Post-Quantum Readiness on VCF (example of industry quantum readiness journey): blogs.vmware.com
NASDAQ / NetApp Press Release — NetApp ONTAP Becomes First Enterprise Storage Platform to Receive NSA Validation: nasdaq.com
NetApp Trust Centre — Commercial Solutions for Classified (CSfC) Program & ONTAP: netapp.com/csfc
NetApp — Autonomous Ransomware Protection: netapp.com/arp
NetApp ONTAP Docs — Autonomous Ransomware Protection (ARP): docs.netapp.com
BusinessWire — NetApp Fights Ransomware in Real Time with Built-In AI: businesswire.com
CISA — Secure by Design Pledge Signatories: cisa.gov
NetApp Newsroom — NetApp Earns AAA Rating for Industry-First AI-Driven On-Box Ransomware Detection Solution (SE Labs, June 2024): netapp.com/newsroom
NetApp — NetApp Wins SE Labs Award for 99% Effective AI Ransomware Protection (2025): netapp.com/arp
NetApp Blog — NIST Cybersecurity Framework: Data Protection and Security: netapp.com/blog/nist
NetApp Newsroom — NetApp Sets New Standard for Cybersecurity at the Storage Layer: netapp.com/newsroom

⚠ Personal Blog Disclaimer The content of this article represents my personal views and opinions only. It is written in a personal capacity and does not represent the views, positions, or endorsements of my employer, clients, or any affiliated organisations. All product names, trademarks, certifications, and brand names mentioned are the property of their respective owners and are referenced for informational purposes only. While every effort has been made to ensure accuracy, readers are encouraged to independently verify information using the references provided. This article is not sponsored, commissioned, or approved by any vendor mentioned herein.

Found this useful? Share it.

If this perspective on storage security resonated, pass it on to a colleague or fellow IT professional.

Explore the references →

Harvest Speaks: Turning ONTAP Metrics Into Conversations

2026-07-14T09:00:00.000+08:00

Storage • Observability • AI Agents

Harvest Speaks: Turning ONTAP Metrics Into Conversations

How an old open-source workhorse quietly became an AI-ready observability layer — and what that means for the way we work with storage data.

⏰ 6 min read

The views expressed here are my own personal opinions, based on publicly available information. They do not represent the views of my employer or any vendor, and should not be taken as official guidance, product roadmap, or endorsement.

There's a particular kind of open-source tool that never quite gets the spotlight it deserves. It doesn't have a flashy launch event. It doesn't trend on social media. It just sits there, quietly doing its job, one poller cycle at a time — until one day you realise half your monitoring stack has been leaning on it for years without you thinking twice.

For a lot of ONTAP and StorageGRID admins, that tool is NetApp Harvest. And this year, it picked up a capability that's worth a proper look: the ability to talk directly to AI agents.

Open Source ONTAP · StorageGRID · E-Series · Cisco Prometheus-native MCP-enabled

What Harvest Actually Is

At its core, Harvest is an open-metrics collector. It reaches into ONTAP clusters, StorageGRID grids, E-Series arrays, and even Cisco Nexus switches, pulls out performance, capacity, and hardware metrics, and routes them into a time-series database of your choice — typically Prometheus or VictoriaMetrics. Grafana then sits on top, turning that data into the dashboards most storage teams are already familiar with.

It's been around long enough to become the default answer whenever someone asks "how do I monitor my ONTAP environment without buying another platform?" It's free, it's actively maintained, and the dashboard library has grown steadily — the most recent major release brought E-Series support to general availability, added dashboards for license tracking and projected time-till-full on aggregates and volumes, and extended coverage to Google Cloud NetApp Volumes and newer Cisco switch models.

The Familiar Architecture

If you've deployed Harvest before, this diagram will look familiar. If you haven't, this is the shape of it:

Nothing about that pipeline changes. Harvest still collects, Prometheus still stores, Grafana still visualises. What's new is that second branch on the right — a direct line from your metrics store into an AI agent.

A Harvest-powered Grafana dashboard, showing cluster-level performance and capacity metrics.

The Pivot: From Dashboards to Dialogue

Dashboards are great when you know what you're looking for. They're less great at 2am when you're trying to figure out, in plain language, "why is this SVM slow" or "which volumes are about to run out of space" — and you don't want to click through six panels to find out.

That's the gap the newest addition to Harvest is aimed at: a Harvest MCP server, built on the Model Context Protocol. Rather than replacing Prometheus or Grafana, it sits on top of the same pipeline you already run, and exposes your metrics through a structured interface that AI agents can query safely and consistently — no PromQL required.

Storage monitoring stops being something you look at, and starts being something you talk to.

MCP-compatible clients — GitHub Copilot, Claude Desktop, and similar tools are the ones NetApp calls out explicitly — can connect to this server and translate natural-language questions into structured queries against your existing metrics store, then hand back the answer in plain language. Your dashboards and alerts don't go away; this just gives you a faster, more conversational way to do the exploratory digging that used to mean opening Grafana and hunting.

Leveraging an Existing AI Agent

Here's the part that matters most for presales conversations and day-to-day ops: if you're already running Harvest, you don't need a new platform to get this. You're extending a pipeline that's probably already in production.

1Keep your existing Harvest pollers and Prometheus (or VictoriaMetrics) deployment exactly as they are. The MCP server reads from what's already being collected — it isn't a second data pipeline.

2Deploy the Harvest MCP server alongside your metrics store, and point it at your existing Prometheus/VictoriaMetrics endpoint.

3Register the MCP server as a connector in the AI agent you already use day to day — Claude Desktop, GitHub Copilot, or any other MCP-compatible client.

4Start asking questions in plain English instead of writing PromQL or opening a dashboard first.

The exact configuration syntax will depend on which AI agent you're using and how you've deployed the MCP server, so I'd point you to NetApp's own Harvest documentation for the precise setup steps rather than me reproducing them here. But structurally, connecting an MCP server to a client like Claude Desktop typically comes down to adding an entry to that client's MCP configuration — something along these lines, illustratively:

{ "mcpServers": { "harvest": { "url": "http://your-harvest-mcp-host:PORT" } } }

Once that connection is live, the AI agent has read access to your Harvest-collected metrics for the questions you ask it — nothing more. It's an additive layer, not a replacement for the governance and access controls you already have around Prometheus and Grafana.

What This Looks Like in Practice

A few examples of the kind of questions this setup is built to answer — the sort of thing that used to mean five minutes of dashboard-hunting:

What's the overall health of my infrastructure right now?

A summarised sweep across clusters, aggregates, and volumes, instead of tabbing between separate Grafana panels.

Which volumes are running out of space?

A direct answer pulled from capacity metrics, without needing to know which dashboard has the right panel.

Which systems are experiencing high latency?

A cross-cluster performance check that would otherwise mean comparing several dashboards side by side.

Why This Matters Beyond the Ops Desk

From where I sit in presales, this is a useful shift to understand — not because it's a new product to pitch, but because it changes the shape of a conversation I have often. Customers with mature ONTAP environments frequently already have Harvest running somewhere. The question used to be "do you need a new monitoring platform to get AI-driven insights out of your storage estate?" Increasingly, the honest answer is: maybe not — you might just need to point the AI agent you already use at the pipeline you already have.

That's a more interesting conversation than a rip-and-replace pitch. It's also a good illustration of a broader pattern worth watching: infrastructure tooling isn't just adding AI features on top, it's exposing itself as data that AI agents can reason over directly, through open standards like MCP rather than proprietary integrations.

A Few Things Worth Keeping in Mind

This is still an evolving area, so a couple of caveats seem worth flagging as you evaluate it. First, an MCP layer is only as good as the metrics feeding it — if your Harvest pollers aren't configured to collect the data you care about, no amount of natural-language querying will conjure it. Second, treat access to an MCP server the way you'd treat any other read path into production monitoring data — with the same network segmentation and access review you'd apply to Grafana or Prometheus itself.

Closing Thought

Harvest has spent years being the tool nobody talks about because it just works. It's a fitting twist that the thing pushing it back into the conversation is, quite literally, conversation. If you're already running it, this might be the smallest lift you'll find all year toward making your storage estate genuinely AI-accessible — no new platform, no rip-and-replace, just a new way of asking the questions you were already asking.

Are you using NetApp Harvest? Share your thoughts and comment.

Your AI is Ready. Your GPU Isn't. Here's the Fix.

2026-06-23T09:00:00.000+08:00

Personal views only. This article reflects my own research and perspective. It is not affiliated with, endorsed by, or written on behalf of any employer or vendor. All sources are publicly available.

When on-premises GPU capacity runs dry, organisations burst workloads to the cloud — only to discover their data can't follow. NetApp FlexCache bridges that gap without moving a single byte you don't need.

🕓 7 min read • AI Infrastructure • Hybrid Cloud • NetApp ONTAP

The GPU queue problem nobody talks about

Enterprise AI adoption is accelerating. Data science teams are running more experiments, training larger models, and processing more inference workloads than ever before. The on-premises infrastructure that was sized for last year's ambitions is now permanently oversubscribed.

The result is the GPU queue — that invisible backlog where jobs sit waiting for compute cycles that simply aren't available. Teams plan their sprints around it. Data scientists schedule notebook runs overnight. Model retraining pipelines stretch from hours into days.

AI workloads are languishing in large queues

Many organisations have given up on cloud bursting altogether because of the data movement challenge — leaving compute capacity untapped and project timelines stretched.

Source: NetApp Community, Tech ONTAP Blogs, 2025

The instinctive answer is to burst AI workloads to cloud GPU instances — AWS, Azure, or Google Cloud all offer elastic, high-performance GPU compute on demand. The economics make sense. The agility is compelling. But there is one problem that stops most organisations before they start.

Their data is on-premises. All of it. And moving petabytes of training data, feature stores, and model artefacts to the cloud — just to run a job — is neither practical, affordable, nor in many cases compliant.

The data gravity trap

Data gravity is the phenomenon where applications and services are drawn towards large concentrations of data — because moving the data is expensive, slow, and risky. For enterprise AI, this is particularly acute.

Consider a typical scenario: a financial institution holds a multi-petabyte data lake on-premises. Regulatory requirements mandate that raw data remains within jurisdictional boundaries. The data science team wants to burst a model training job to cloud GPU instances to avoid a three-week queue. To do so with conventional approaches, they would need to:

Transfer the dataset

Move potentially hundreds of terabytes over a WAN connection, consuming bandwidth and time.

Manage consistency

Ensure the cloud copy stays synchronised with the on-premises source as data changes.

Satisfy compliance

Demonstrate that sensitive data has not left regulated boundaries or been replicated without authorisation.

Clean up afterwards

Track and delete cloud copies to avoid creating uncontrolled data sprawl or residual compliance exposure.

It is no wonder many organisations simply abandon the idea and endure the queue. The overhead of managing a full data copy in the cloud can easily outweigh the benefit of the elastic compute.

What is NetApp FlexCache — and why is it different?

NetApp FlexCache is not a replication technology. It is not a copy. It is a sparse, intelligent cache that makes remote data appear to be local — without physically moving it until the moment it is actually needed.

“A FlexCache volume is an intelligent sparse container of the source data, which looks and feels the same as the source data to the application.”
— NetApp Documentation

Here is how it works in a hybrid AI context:

The FlexCache volume is created in the cloud

A cache volume is provisioned on Amazon FSx for NetApp ONTAP, Azure NetApp Files, or Google Cloud NetApp Volumes. It is linked to the on-premises ONTAP origin volume. At this point, no data has moved. The cloud volume appears to contain the full namespace of the source — every file, every directory — but the actual data blocks are not yet present.

Cloud GPU instances read from the FlexCache

Your AI training job starts. When the cloud compute node requests a data block, FlexCache first checks whether that block exists in the local cache. If not (a cache miss), it fetches only that specific block from the on-premises origin — not the entire file, and certainly not the entire volume.

The working dataset builds up naturally

Blocks that are accessed are cached in the cloud environment. Subsequent reads to those same blocks are served at cloud-local speed — no WAN round-trip required. Over time, the cache becomes warm with exactly the data the workload actually uses, and nothing more.

Writes return to the origin automatically

When the AI job produces output — model checkpoints, inference results, processed datasets — those writes pass through the FlexCache back to the on-premises origin volume. The cloud never becomes the source of truth. Your data governance posture is maintained throughout.

Delete the cache. Leave no trace.

When the job is complete, delete the FlexCache volume. There are no untracked data copies, no residual data in cloud storage, and no ongoing management overhead. The origin on-premises remains the single, authoritative copy.

The compliance advantage: just enough, for just long enough

For regulated industries — financial services, healthcare, government — the compliance posture of a hybrid AI pipeline is often the deciding factor between adoption and stagnation.

FlexCache changes the compliance conversation fundamentally. Rather than arguing about whether a full copy of sensitive data can reside in a cloud region, the question becomes: can transiently cached blocks, used solely for computation and automatically deleted on job completion, satisfy the relevant regulatory framework? In many cases, that is a far easier case to make.

✕ Traditional copy approach

Full dataset replicated to cloud
Ongoing synchronisation required
Data residency risk for every byte
Cleanup is manual and error-prone
Storage costs scale with dataset size

✓ FlexCache approach

Only accessed blocks move to cloud
Cache coherence managed automatically
Source of truth remains on-premises
Delete cache = zero residual data
Storage costs scale with working set

Additionally, because management and data protection are implemented at the origin volume only, your DR strategy, encryption posture, and access controls do not need to be duplicated in the cloud. Governance remains centralised even as compute becomes distributed.

Where this matters most in enterprise AI

Model training bursts

Training runs that exceed on-premises GPU capacity burst to cloud, accessing only the training batch data they need rather than the full corpus.

Inference at scale

High-throughput inference workloads access feature stores and model artefacts stored on-premises, served via cache to cloud GPU pools without wholesale data migration.

EDA and simulation

Design files stored on-premises are accessed at cloud-local speeds by burst compute clusters, eliminating WAN latency without replicating the entire design library.

Multi-cloud data access

FlexCache removes cloud silos by enabling data from one cloud region or provider to be cached efficiently in another, supporting distributed AI pipelines.

Where FlexCache fits in a broader AI data strategy

At NetApp Insight 2025, CEO George Kurian outlined a four-layer model for enterprise AI infrastructure: data infrastructure modernisation, an AI data pipeline, cloud transformation, and cyber resilience. FlexCache sits squarely in the cloud transformation layer — as the mechanism that makes hybrid cloud AI workloads practical rather than theoretical.

This is not an isolated feature. FlexCache is part of a broader ONTAP capability set that includes the global namespace, multi-protocol support (NFS and SMB/CIFS), and integrated encryption — all of which mean AI workloads can access cached data without re-architecting pipelines or re-engineering storage access patterns.

Key capabilities

Global file system namespace

The cache presents the same namespace as the origin — no path remapping required.

Disconnected mode

If the WAN link drops, already-cached blocks remain accessible — the job does not fail.

Cache pre-warming

Pre-populate the cache before a scheduled job to eliminate cold-start latency.

Encryption compatibility

Encrypted origin volumes are fully supported, integrated with cloud KMS services on AWS and Azure.

You only pay for what you actually use

One of the most underappreciated benefits of FlexCache is not about latency or architecture — it is about cost. To understand why, you need to understand the concept of the working set.

In any AI training or inference job, the workload does not touch the entire dataset. It accesses a specific slice: the current batch, the active feature columns, the relevant time window. This active slice — the data the job actually reads during its run — is the working set. For a large enterprise data lake, the working set for a single job might be 2% to 10% of the total dataset volume.

With a traditional copy-to-cloud approach, you pay for the full dataset. Every terabyte, whether the job touches it or not. With FlexCache, you pay only for the blocks that are actually accessed — because those are the only blocks that ever arrive in the cloud cache. As your datasets grow, the gap between the two approaches widens dramatically.

Cloud storage cost: full copy vs FlexCache working set (illustrative)

Illustrative. Actual savings depend on working set size relative to total dataset volume. ↑ indicates cost exceeds chart scale.

This dynamic changes the economics of hybrid AI fundamentally. A team with a 50 TB genomics data lake does not need to provision and pay for 50 TB of cloud storage to run a burst training job. If that job's working set is 800 GB, that is all that ever lands in the cloud cache — and when the job ends and the cache is deleted, it is all gone. No residual storage bill, no orphaned data.

Going further: when an AI agent manages the burst itself

The hybrid bursting pattern described here — detect GPU saturation, provision a FlexCache volume in the cloud, run the job, clean up — is a well-defined workflow. And well-defined workflows are exactly what agentic AI is designed to automate.

🤖

Agentic AI in action

NetApp's published research describes an AI agent — powered by a large language model — that monitors on-premises GPU queue depth in real time. When utilisation crosses a defined threshold, the agent autonomously provisions cloud GPU instances, creates a FlexCache volume linked to the on-premises origin, submits the queued workload to the cloud cluster, and tears down the cache and compute once the job completes. No human intervention. No manual handoff.

This is not a future concept. It is a documented reference architecture available on the NetApp Community blog. For the full technical walkthrough, see: “Agentic AI in action: automated cloud bursting when GPU capacity is reached” — Tech ONTAP Blogs, NetApp Community, 2025.

What makes this particularly significant is that FlexCache removes the last friction point from agentic orchestration. An AI agent can automate compute provisioning easily enough — cloud APIs make that straightforward. The harder problem has always been data: how does the agent ensure the cloud compute can actually access the right data without triggering a manual data migration? FlexCache answers that question, making fully autonomous hybrid AI bursting a realistic operational model rather than a theoretical one.

Further FlexCache capabilities worth knowing

Beyond the core bursting pattern, two additional FlexCache behaviours are particularly relevant for enterprise AI workloads operating in hybrid environments.

🔗 Disconnected resilience

If the WAN link between on-premises and cloud is interrupted mid-job, FlexCache does not fail the workload. Any blocks already present in the cloud cache remain fully accessible, allowing the job to continue against the data it has already pulled — without requiring a live connection to the origin.

For long-running training jobs where network reliability cannot be guaranteed, this is a meaningful safety net. The job does not have to restart from scratch because of a transient connectivity event.

⚡ Cache pre-warming

FlexCache supports pre-population of the cache volume before a scheduled job begins. Rather than incurring cold-start latency — where the first pass through the dataset is slow because every block is a cache miss — teams can warm the cache in advance with the expected working set.

For time-sensitive jobs or pipelines with predictable data access patterns, pre-warming ensures that cloud GPU instances hit the ground running at local-cache speeds from the first read.

FlexCache capability summary

Demand-based block caching Write-back to origin Disconnected resilience Cache pre-warming Global namespace NFS & SMB/CIFS support Zero residual data on deletion

The question is no longer whether to burst — it is how

Cloud GPU elasticity exists. The economics are compelling. The only legitimate barrier has been data — specifically, the cost, complexity, and compliance risk of moving it.

FlexCache reframes that barrier entirely. Data does not move to the cloud — access to data moves to the cloud. The distinction sounds subtle but the practical implications are significant: no bandwidth cost for data you never access, no compliance exposure for data that never leaves the origin, no management overhead for copies that cease to exist the moment the job is done.

For enterprise AI teams sitting behind a GPU queue, that is not a minor optimisation. It is the difference between a viable hybrid cloud AI strategy and one that stays on the whiteboard.

Want to go deeper on hybrid AI infrastructure?

Explore NetApp's published documentation on FlexCache, FSx for ONTAP, and the AI Data Engine — all publicly available.

FlexCache docs → FSx caching guide →

Disclaimer: This article represents my personal views and independent research only. It is not written on behalf of, endorsed by, or affiliated with any current or former employer, or with any vendor referenced herein. All technical claims are based on publicly available documentation and sources linked where referenced. Readers are encouraged to verify all information independently before making infrastructure decisions.

Garbage in, AI out. Why 60% of AI projects are set to fail.

2026-06-02T09:00:00.000+08:00

Data & AI Strategy

Garbage in, AI out.
Why 60% of AI projects are set to fail.

And what you can do to be in the 40% that succeed.

⏱ 5 min read Artificial Intelligence Data Management NetApp Enterprise AI

Personal views disclaimer: The opinions expressed in this article are entirely my own and do not represent the views, positions, or official communications of my employer or any organisation I am affiliated with. References to third-party products and companies are based on publicly available information and my personal professional observations.

Bad data doesn’t get better when processed. It just gets more elaborate — and at Agentic AI scale, it multiplies.

At NetApp Insight 2025, this image was paired with a Gartner prediction that I haven’t been able to stop thinking about since.

60%

of AI projects will be abandoned through 2026 due to lack of AI-ready data.

— Gartner, Roxane Edjlali, Senior Director Analyst, February 2025

The two reinforce each other perfectly, and together they tell us something that many enterprises are still reluctant to hear.

AI adoption is accelerating — faster than any technology before it

AI is now the go-to initiative for virtually every enterprise. Adoption is rising faster than any technology in recent history, driven not just by investment pressure but by a genuine need to stay competitive. If you are not moving, you will eventually fall behind.

 Bank A — with AI

Loan application initial approval in hours

⏳ Bank B — manual process

Same application takes days

That gap is not just an efficiency difference — it represents real business opportunity won and lost.

The uncomfortable truth: data is what’s missing

With heightened awareness of AI, many organisations have placed it at the centre of their core initiatives. Some have already completed proof-of-concepts or started adoption. These investments are significant, and yet Gartner predicts that 60% of AI projects will be abandoned through 2026 — not because of the wrong model, not because of insufficient compute, but because the data was not ready.

“We can have the most advanced AI infrastructure, top-notch software, applications, and specialists — and still fail. The root cause is always data.”

In the rush to adopt AI, it is the most critical ingredient that gets overlooked. Organisations prioritise the toolset and forget the foundation.

Data: the heart of everything

I have had many meaningful conversations with customers about this very issue. What consistently surprises people is how often data readiness is assumed rather than verified. The image at the top of this post captures it with brutal simplicity — it does not matter how sophisticated your AI layer is. If the data going in is poor quality, the output will be too. Agentic AI simply means bad data, now operating at scale.

This is where NetApp enters the conversation. NetApp has been in the data management space for over three decades. Since AI is largely about unstructured data — files, images, video, audio — their focus on data management, security, protection, availability, optimisation, and accuracy is directly relevant.

What good AI data engineering looks like

Effective AI data engineering means understanding where data lives, keeping it current without unnecessary duplication, enforcing security and classification, and transforming data into a form readily consumable by AI applications. Doing this natively would typically require stitching together a dozen or more different tools and platforms — a level of complexity that stalls most AI initiatives before they reach production.

From what I have seen at NetApp Insight 2025, the NetApp AI Data Engine (AIDE) is one approach that addresses this — making data globally available without duplication across silos, underpinned by ONTAP’s enterprise data mobility and protection capabilities.

Before data is ready for Enterprise AI, the right foundations must be in place — starting with Data Classification. NetApp’s Data Classification service provides over 200 out-of-the-box classifications to identify sensitive information within files. Watch the ONTAP Data Classification demo to see it in action.

Three things to get right for AI success

Standardise your infrastructure

Avoid the rabbit hole of certifying every component of a three-tier architecture for AI. Choose a validated design and start there. Validated reference architectures like Cisco FlexPod with NetApp are a good example of a proven, AI-ready starting point.

Acknowledge tool complexity

No single vendor covers every tool needed for data engineering. A dozen or more tools are typically required, and a lack of internal capacity to manage that complexity will stall AI projects — often quietly and expensively.

Focus on Enterprise AI — not the plumbing

Make your data AI-ready without getting lost in architectural engineering. Leverage platforms and partner ecosystems — such as the NetApp AI ISV partner ecosystem — to offload data engineering tasks so your team can focus on value, not infrastructure.

The ecosystem that completes the picture

While many sessions and articles cover AI deployment and security, very few address how data should be engineered and prepared securely beforehand. Most assume enterprises have already handled this. As a data owner, you need control over access without needing to master the entire AI stack.

Three decades of data management expertise means platforms like NetApp understand where data is, how to keep it current, and how to make it secure and AI-ready. The AI Data Engine is further complemented by technology partners like Diskover, which discovers data in media formats and feeds that intelligence into the AIDE metadata architecture — without copying the data itself.

40%

of AI projects will succeed. Will yours be one of them?

Gartner’s prediction is a challenge, not a verdict. The organisations that succeed will have done the unglamorous work of preparing their data properly. Reach out to your NetApp representative or partner to find out how to get started.

Disclaimer: This article reflects my personal opinions and professional observations only. It does not represent the views or official positions of my employer or any affiliated organisation. All product references are based on publicly available information. Product and company names mentioned are trademarks of their respective owners.

References

1. Gartner — “Lack of AI-Ready Data Puts AI Projects at Risk,” Roxane Edjlali, Senior Director Analyst, February 26, 2025. gartner.com ↗
2. NetApp Data Classification — netapp.com/data-services/classification ↗
3. ONTAP Data Classification Demo — youtube.com ↗
4. NetApp Insight 2025 — “Unleashing GenAI on ONTAP: First E2E AI Data Engine Customer Deployment.” netapp.com ↗
5. NetApp AI ISV Partners — netapp.com ↗

Beyond the Specs: Pre-Sales Architect Value in the Era of Flash

2026-04-21T09:00:00.002+08:00

Recently I met with a customer who was genuinely frustrated. Their monitoring tools were showing massive, unexplained spikes in utilization across their storage controllers. Naturally, the immediate assumption was that the hardware was failing or undersized. They were ready to throw more money at a storage "problem" that didn't actually exist.

It wasn't until we sat down in person and I dug into their VDI (Virtual Desktop Infrastructure) environment that the truth came out. The hardware was fine; the best practices were missing.

As Solution Architects and Consultants, our job isn't just to sell boxes—it’s to educate. We need to ensure customers don't misidentify software misconfigurations as hardware limitations. In the world of VDI, performance is often a matter of "prudent design" rather than just "faster disks."

The "Storm" Before the Calm: Managing the Boot Storm

Think of your VDI environment like a highway at rush hour. When hundreds of users log in between 8:00 AM and 9:00 AM, they create a Boot Storm. Every virtual machine is trying to pull data from the storage controllers simultaneously.

The Fix: Use pre-provisioning. By setting your desktop pools to spin up VMs before the shift starts, you spread the storage load over a longer window, effectively smoothing out the traffic spike before the first employee even touches their keyboard.

Policy Optimizations: Trimming the Fat

Beyond hardware, look at your Citrix or VMware policies. Large user profiles can choke a login process.

Profile Management: Use tools like Citrix Profile Management or folder redirection. By redirecting heavy folders (like Documents or Downloads) to a separate network share, you reduce the amount of data that needs to be "copied" to the VM during login.

GPO Cleanup: Audit your Group Policy Objects. Complex login scripts, unnecessary drive mappings, and printer searches add seconds to every login and IOPS to every controller.

Antivirus & Backups: Don't Let Security Slow You Down

If not managed properly, security software can be its own worst enemy.

AV Best Practices: Scan your master image thoroughly before deployment, then set exceptions for VDI-specific files. Schedule scans to run in "staggered" windows during off-peak hours so they don't all hit the controller at once.

Smart Backups: Avoid backing up data from inside each individual VM. Use folder redirection to a central share and back *that* up. It’s more efficient and keeps your VDI storage focused on performance, not data protection tasks.

Key Takeaways for the Modern IT Leader

Utilization ≠ Capacity: High controller usage often signals a bottleneck in process, not a lack of hardware.

Holistic Solutioning: In today’s Flash-heavy world, we must be more prudent in sizing. Flash is fast, but it isn't a "magic wand" for bad configuration.
Education is Performance: A well-educated team that follows best practices will always get more ROI out of their storage than a team that simply over-provisions.

Final Thought:

Prudent sizing isn't just about meeting a budget; it's about precision. As architects, we bring the most value when we help customers see that the best solution isn't always a bigger hammer—it’s a better blueprint.

Will AI know this, maybe not before. But now, it does!

Storage in a Seller’s Market: Flash Prices, Licensing Shifts, and Why IT Leaders Must Move Faster

2026-03-02T08:48:00.031+08:00

Disclaimer as always: Opinions all solely my own.

The storage industry is entering a phase that feels uncomfortably familiar: constrained supply, rising prices, and vendors tightening control over how customers consume technology. Two seemingly separate forces are converging to reshape IT decision-making:

The NAND flash supply crunch and price escalation
Post-acquisition licensing shifts following Broadcom’s acquisition of VMware

Together, they signal a transition from a buyer’s market to a seller-advantaged landscape—one where hesitation, legacy assumptions, and slow procurement cycles can carry real financial and operational penalties.

The VMware Effect: When Licensing Strategy Becomes a Market Signal

The acquisition of VMware by Broadcom marked more than a corporate transaction—it signaled a shift in how enterprise software value is captured.

VMware’s move from perpetual licensing to subscription models, and the bundling of core capabilities into VMware Cloud Foundation (VCF), changed the negotiation dynamic. Customers who once purchased only what they needed—such as vSphere—now face bundled licensing that increases total cost and reduces flexibility.

Why Customers Felt Cornered

It’s easy to say, “Just don’t renew.” In reality, most enterprises couldn’t:

Migration timelines: Platform assessments and migrations take years, not months.
Application certification: Many production systems are certified only on VMware platforms.
Compliance requirements: Running unsupported infrastructure is not acceptable in regulated environments.
Operational risk: Downtime or instability costs far more than licensing increases.

This isn’t traditional lock-in—it’s dependency without viable short-term alternatives. VMware’s technology remains mature and excellent. The shift is in commercial posture, not technical merit.

The lesson for infrastructure leaders:

Vendor strategy can change faster than your ability to migrate.

NAND Flash Shortage: Déjà Vu With Higher Stakes

At the same time, the industry is confronting a surge in NAND flash demand, driven by AI workloads, hyperscale expansion, and supply constraints.

What’s Happening on the Ground

Server and storage hardware price increases reported up to 70% in some cases.
Quote validity shrinking from 30 days → 2 weeks → 24 hours.
Vendors rationing supply and prioritizing high-value customers.
Consumer impact likely to follow (laptops, phones, TVs, and smart devices)

This is no longer a theoretical supply issue—it’s operational reality.

Vendor Responses: Different Speeds, Same Direction

Some vendors are increasing prices gradually; others have implemented sudden spikes. For example, companies like NetApp appear to be pacing increases more cautiously—but with global stockpiles shrinking and flash prices trending upward, gradualism may be temporary.

No vendor is immune to upstream component costs.

Rethinking Hyperconverged Infrastructure in a High-Cost Era

Hyperconverged infrastructure (HCI) has been a fantastic solution for simplifying operations and scaling predictably. But in today’s climate, it’s worth reassessing whether it remains the most cost-effective approach.

HCI tightly couples compute, storage, and memory—meaning scaling storage often requires buying more servers and expensive RAM, even when compute isn’t needed.

Vendors like Nutanix built their early momentum on a strong “No SAN” message—positioning HCI as the end of external storage. That message resonated in an era of falling flash prices and abundant hardware supply.

In a market where:

Server prices are rising
Memory costs remain high
Flash prices are surging
AI demanding large capacity and sustained performance

AI introduces a new challenge: large datasets, high-throughput pipelines, and GPU-driven processing require storage architectures that scale capacity and performance independently. In many cases, HCI struggles to meet these demands efficiently because:

Scaling storage requires adding compute nodes
GPU workloads need dense, high-performance shared storage
Large AI datasets exceed the cost-effective limits of node-based scaling

We are seeing the industry narrative evolve. The strict “No SAN” stance is softening as customers demand greater flexibility, and the conversation increasingly includes external storage integration and converged models.

We’re seeing renewed interest in converged infrastructure—separating compute and storage—to allow:

Independent scaling of storage without buying more servers
Better cost control when flash prices spike
High-performance shared storage better supports AI pipelines
Flexibility to adopt hybrid storage tiers

HCI is still the right choice for many workloads. But the assumption that it is always the most efficient architecture no longer holds. Cost dynamics have changed.

Hybrid Strategies Beyond Tering: Cloud as a Pressure Value

Tiering between flash and HDD is one lever. Another increasingly relevant option is hybrid cloud as an interim capacity strategy.

When on-prem flash costs spike or supply tightens, the ability to extend workloads or datasets into the cloud—then migrate back when conditions stabilize—can provide critical flexibility.

This approach allows organizations to:

⁠Avoid locking into high on-prem flash costs during shortages
⁠Scale capacity temporarily without large capital outlays
⁠Maintain performance by placing the right workloads in the right location
⁠Repatriate data when flash pricing normalizes

The key is seamless mobility. Solutions that enable consistent data management, replication, and workload portability across on-prem and cloud environments become strategic—not optional.

Without this capability, hybrid cloud becomes a one-way door. With it, it becomes a financial and operational safety valve.

The Illusion of “Better Deals” in a Shortage

When shortages hit, customers naturally explore alternatives.

Option 1: New Supply Chains (e.g., Chinese Vendors)

Vendors such as Huawei, Inspur may appear attractive due to different supply channels. But considerations include:

Support language and regional coverage
Geopolitical and compliance risks
Supply prioritization (domestic vs. international customers)

Option 2: Vendors Holding Prices

Some flash-only vendors are not raising prices—for now.

But ask the hard questions:

How long can they absorb rising costs?
Will delayed increases become more severe later?
If they offer only flash, what happens when you must scale?

If your architecture has no HDD tier, your choices become binary: pay or stop growing.

Flash-Only Architectures: The New Lock-In Risk

The industry spent a decade promoting “all-flash everything,” especially as flash approached HDD price parity. That parity is now eroding due to AI demand and supply constraints.

Flash-only vendors—particularly those born in the all-flash era—leave customers with no cost-optimization path when prices surge.

No HDD tier. No hybrid option. No escape valve.

That is a different kind of lock-in: economic lock-in.

Rethinking Optimization: Hybrid Is Back

What was once considered legacy is becoming strategic again.

Cost-Control Strategies Regaining Relevance

Flash as cache, HDD as capacity tier
Automated tiering between flash and disk
Workload placement based on performance needs
Shorter planning horizons (months, not years)

These are not the most glamorous architectures—but they may be the most survivable in today’s market.

Trade-offs to Acknowledge

HDD increases power usage and carbon footprint
Performance tuning becomes more complex
Not all workloads tier cleanly

Still, optimization today is about resilience, not elegance.

Procurement Must Evolve: Speed Is Now a Competitive Advantage

One of the most overlooked inefficiencies is the traditional procurement cycle.

The Old Model

Multi-month evaluations
Extensive RFP cycles
Year-long budget buffering (10–20%)

The New Reality

Prices change in weeks—or days
Supply disappears overnight
Delayed decisions cost real money

Organizations that can decide and execute quickly secure better pricing and guaranteed supply.

This is not recklessness—it is operational agility.

Key Considerations to Avoid Future Lock-In

When evaluating storage solutions in today’s environment, focus on:

1. Architectural Flexibility

Can you mix flash and HDD?
Can workloads move across tiers?

2. Supply Chain Resilience

Where do components originate?
How does the vendor prioritize shortages?

3. Licensing and Commercial Model

Are bundles forcing unwanted capabilities?
What is the exit cost?

4. Scaling Economics

What happens when you need 2× capacity?
Are there lower-cost expansion paths?

5. Procurement Agility

Can you approve purchases quickly?
Are governance processes slowing response?

The Bigger Shift: From Buyer’s Market to Seller’s Market

For years, customers held the advantage: abundant supply, declining flash costs, and flexible licensing.

That era is ending.

Today:

Vendors control supply.
Licensing models consolidate value.
Flash prices are rising.
Alternatives require time most organizations don’t have.

The new mandate for IT leaders is clear:

Make decisive, informed choices that preserve optionality.

Because in a seller’s market, the greatest risk isn’t paying more today—it’s losing the ability to choose tomorrow.

If you’re seeing similar trends in your environment, I’d value your perspective. How are you adapting your storage architecture and procurement strategy in this new climate?

The New Mandatory RFP Questions

2026-02-17T09:30:00.000+08:00

Customers are no longer asking if you can deliver; they are asking how you secure the components to do so. In coming months, we are likely to see the following requirements become standard:

Proof of Component Allocation: Buyers are demanding evidence that the vendor has a direct "Line of Sight" to the silicon. This is where software-only vendors struggle; they can only provide a quote from an OEM, whereas a hardware-owning vendor can provide a production slot.
Price Lock Guarantees: Given the 90% surge in DRAM prices, RFPs now frequently include clauses requiring price protection for 6 to 12 months. Vendors who buy through a middleman (OEM) often cannot commit to this, as they have no control over the OEM’s sudden price adjustments.
Alternative Component Validation: Customers are asking for pre-validated designs that use multiple types of flash or memory modules. If a specific module hits a shortage, can the vendor pivot without requiring a 3-month re-certification process?

Why Transparency is the New "High Availability"
In the past, we designed for High Availability (HA) to survive a controller failure. Today, we must design for Supply Chain HA to survive a manufacturing failure.

Requirement	Software-Only Vendor (OEM Dependent)	Hardware-Owning Vendor (Direct)
Priority	Secondary (subject to OEM internal demand)	Primary (internal control of factory lines)
Pricing	Passive (receives price hikes from OEM)	Active (direct negotiation with foundries)
Visibility	Opaque (blind to OEM component stock)	Transparent (real-time manufacturing tracking)

The "Sovereignty of Hardware"
As a former VMware specialist, I spent years championing the "Software-Defined Everything" movement. However, the current shortage has taught us a humbling lesson: Software cannot run on a slide deck.

If a NEO cloud provider or a large enterprise wins a massive contract, they need to know that the 50PB of storage they ordered will arrive on Tuesday—not "whenever the OEM fulfills the software vendor's order." This is why transparency is becoming a safeguard against "Ghost Capacity"—storage that has been bought and paid for but cannot be delivered.

Architecting for the future now requires auditing the supply chain. In 2026, the best software in the world is a liability if the vendor doesn't own the "metal" it runs on.

A checklist for your reference:

The Architect’s Checklist: Evaluating Supply Chain Resilience
When designing your next-generation data center or cloud expansion, use these questions to help assess the resilience of the hardware supply chain.[ ] Component Sourcing: Is there clarity on the origin and sourcing of key components?

[ ] Supply Chain Visibility: Is there transparency into the different stages of the supply chain?
[ ] Risk Mitigation: Are there strategies in place to mitigate potential disruptions in the supply chain?
[ ] Lead-Time Communication: How is information about potential lead times communicated?
[ ] Scalability Support: How is future scaling of compute or storage addressed in terms of hardware availability?
[ ] Component Flexibility: Is there flexibility in the types of components that can be used if specific parts become unavailable?

Considering the hardware supply chain is an important part of designing resilient IT infrastructure. Even the most sophisticated software can be impacted if the underlying hardware is not readily available.
In 2026, a truly resilient design includes a thoughtful consideration of the hardware's journey from manufacturing to deployment.

The "Software-Defined" Trap: Why Your Architecture is Only as Good as Your Vendor’s Warehouse

2026-02-05T10:20:00.002+08:00

In my years transitioning from the virtualization front lines at VMware to my current role at NetApp, I’ve seen my share of "once-in-a-generation" market shifts. But the current landscape of 2026 is truly unprecedented. As reported by The Wall Street Journal, we are witnessing a "permanent reallocation" of manufacturing capacity toward AI, triggering a memory-chip shortage that industry veterans call the "craziest time ever".

For architects and IT leaders, this isn't just about rising costs; it's a fundamental challenge to how we design for resilience and delivery.

The 2026 Context: A Market in "Hyper-Bull" Phase
The data is stark. Prices for memory surged 50% in late 2025, but that was just the preamble. According to the latest TrendForce projections, contract prices for conventional server DRAM are set to skyrocket by 90% to 95% in Q1 2026 alone. We aren’t just looking at budget adjustments; we are witnessing a complete decoupling of price from value.

This "super-cycle" is driven by an insatiable appetite for High-Bandwidth Memory (HBM) to fuel AI infrastructure. Major suppliers like Samsung, SK Hynix, and Micron are prioritizing these high-margin AI orders, leaving a massive supply gap for conventional server DRAM and NAND drives.

The Hidden Risk: Principal Vendors vs. Hardware Ownership
In this environment, a vendor's "hardware DNA" becomes a critical architectural consideration. Modern "software-defined" principal vendors—think VAST Data, Weka, or Qumulo—offer incredible innovation, but they often lack their own manufacturing lines or deep-tier hardware supply chains. Instead, they rely heavily on OEM hardware from giants like Dell, HPE, or Supermicro.

This creates a dangerous dependency in a shortage:

The Prioritisation Problem: If a hardware giant like Dell has its own multi-million dollar project and a third-party software vendor also needs those same servers for a client, who do you think gets the allocation? In a world where Dell's COO recently stated they have "never witnessed costs escalating at the current pace," internal projects will almost always take precedence.
The "Hidden" Price Gouging: Perhaps more concerning is that software-only vendors have zero control over the hardware pricing set by their OEM partners. An OEM may price their hardware much more aggressively for their own internal projects while charging third-party software partners a premium. Consequently, customers may unknowingly be paying sky-rocketing prices because their chosen vendor is at the mercy of another company's price list.
Delivery Deadlines: Customers considering solutions from vendors who do not own their hardware face a high risk of "slipping" timelines. For a NEO cloud provider or an enterprise needing to scale capacity quickly, a six-month delay isn't just an inconvenience—it’s a threat to the business going live.
Pricing Volatility: Third-party software vendors have little leverage. Since writing, TrendForce has upgraded its outlook. Conventional DRAM contract prices are now projected to surge by 90% to 95% in Q1 2026 (nearly double). These costs are passed directly to the customer, often after the initial quote.

The Architect’s Mandate: Designing for Stability
As architects, we’re taught to design for performance and scale. But in 2026, we must also design for availability of the physical asset.

Choosing vendors like NetApp, Dell, or HPE—who own and manage their own hardware supply chains—minimizes these dependencies. NetApp, for instance, manages the entire stack from the ONTAP software to the physical controller. These "full-stack" vendors have the scale to secure long-term supply agreements and the manufacturing priority to ensure that when you hit "order," the hardware actually exists.

A design is only as good as its delivery. If your solution is held hostage by an OEM's priority list, the architecture has failed. We must prioritize vendors who can guarantee the timeline alongside the technology.

NetApp Shift Toolkit - Migration to Alternatives Made Too Easy

2026-01-06T22:30:00.015+08:00

The discussion on moving to an alternative hypervisor due to Broadcom acquisition of VMware has been a popular topic mainly due to the cost increase of the licenses required to run VMware software. But let's be honest, there are applications that works only on VMware due to certification or qualification. There are still a bulk of workloads that doesn't requires and this is where an alternative is helpful to drive the cost down.

Even if you have managed to assess and decided on the alternative choice (whatever that is), there are just one important cost that many has overlook and it is the most important. That is migration cost from VMware data format to your alternative data format.

There many paid software for such requirements which are not free and generally price per VM or by the amount of data capacity. They are reliable but purchasing the additional cost for a one time use might not be as viable, not to mention the professional service cost needed on top of that. This takes up cost and time.

The other method would be to use an existing backup software that you own that support VMware and the alternatives, this allows you to backup from one and restore on the other. This save cost but not on time.

Both solutions are all good choices, but both have a common issues; Time. How long can you have your infrastructure keep down for migration to be completed, tested and ready for production again, not mentioning the needed reconfiguration if needed.

That is where NetApp introduce NetApp Shift Tookit. Regardless the size of the VMs, you can do migrate/convert them in minutes. Thanks to Ahmed Al-Nabhani for providing the the details for me to provide this article.

How is that possible where other software is unable to do that?

The magic lies on the storage itself. NetApp as a Storage provider leverage on its storage platform technology to perform the miracles on this. Read this on how it does its magic.

NetApp Shift Toolkit not only reduce the amount of time needed by typical migrations tools on the market, the best part is it comes free on all NetApp storage.

Customers can utilize this on their own, or engage NetApp or its NetApp partners to provide the professional services for assistance.

What are the alternatives hypervisor are supported?

In short, VMware bidirectional to and from Hyper-V or VMware to any KVM variant that support qcow2 format. I would say almost all KVM hypervisors that support external storage. Check this overview page.

For those interested in the migration workflow, you can refer to this.

In summary, if. you have a need for such migration, do check out NetApp Shift Toolkit. Not only it help in cost but also in time. Which is the most costly variable for an alternative migration.

At the meantime, if you are running any virtualization workload be it on VMware, Hyper-V or any KVM alternatives, NetApp is the only vendor that support almost all the KVM variants and NetApp ASA is only Block storage giving you the ONTAP experience specially catered for virtualization and databases workload with its small footprint yet powerful with a good 99.9999% service level. Check out the details here.

NetApp ONTAP VScan Technology

2025-12-22T09:30:00.015+08:00

Where I used to come from VMware, it build an ecosystem of integration with its vendors. And have a massive interops from compute, storage and network. It does this through an extensive API within its full software stack.

When I joined NetApp earlier this year, NetApp have a Cyber Resilience Partner Ecosystem surrounding data. One of the category is Virus Detection. This was made possible with ONTAP Vscan technology which is a proprietary to NetApp. Which also means, not all AV vendor can just integrate. They need to be using ONTAP Vscan technology for that integration.

Now the questions is won't that made it less compiling? Maybe, but for the sake of security, this is needed. Why do I say so?

Generally on the market, storage vendors all have integration to AV vendors in some way, However, most choose the "easy way out". They leverage on ICAP (Internet Content Adaptation Protocol). As ICAP is a generic protocol, it can be easily utilized. However, one issue with ICAP is, it is not filtered. Which means it requires additional network traffic sending information whether needed or not for the integration. This can affect performance and latency. Why do you need unneeded data?

And since it utilize ICAP, ICAP will require an additional server as a proxy. Which also results in higher compute resources when demand increases, especially when it scales, the additional traffic will be very demanding. This also due to the unnecessary data sent.

On security, ICAP sends the exact files to the ICAP server. Which means while protecting against virus, we just introduce a new exposure of security vulnerability.

ONTAP Vscan since its native part of the ONTAP system, everything happens within the storage without sending any traffic across the network. There is a ONTAP Antivirus Connector, to handle the request between the AV server and NetApp storage. This not only make it easy to enable but also help to scale much more easily without the complexity.

Here is a short writeup on how ONTAP Vscan works.

In summary, NetApp has been the most secure storage on the planet for a reason. Taking security at it's utmost priority to ensure purposefully addressing the needs instead of fufilling a check box.

What's New at NetApp Insight 2025

2025-10-24T09:00:00.002+08:00

The NetApp Insight is NetApp flagship event which showcasing the new innovation and launches announcement.

This year without fail, NetApp once again has release one of the most anticipated offering in the AI arena. The first in able to do this all right from the start doing it right more than 30 years ago with ONTAP, AFX and AIDE.

Together with some new branding and new abilities. Let's go right into it into some of the major updates.

1. AFX (link) and AIDE (link)

AFX is a new offering which is NetApp first announced in NetApp Insight 2024 and now its available. This is NetApp first disaggregated storage solution cater to demanding workload primary AI but also other such as EDA. The best part is certified with NVIDIA, it not just meet the performance but also inherit all the best of ONTAP such as Security and more.

AIDE, on the other hand, as the name implies, its support the AI Data Pipeline. This is the first and only AI Data Engine that is available on the market. Just imagine the data challenges you will face when starting AI initiatives:

Gathering data from all over such as public cloud, and different locations, provided you know where they are. Knowing where to start.
Answer: With NetApp, you get to know where all your data are located since every data services whether on prem, in the cloud or as a software are running ONTAP.
Making sure data that are gathered remain current. Repeating point 1. all over again.
Answer: With NetApp, leveraging ONTAP SnapDiff to ensure only the data differences are update efficiently. This is often a huge challenge.
Applying classification and guardrail and ensuring data is secured and providing governance adhere to the local policy or guideline.
Answer: NetApp is able to work that all out within AIDE, providing data classification and governance the type guardrail you need on your data in a regardless data distribution as long they are running ONTAP.
Creating dataset used for different AI Agents and chatbots.
Answer: Containing all the data metadata, storing it its own vector database using NVIDIA Inference Microservices (NIM). This vector embeddings efficiently and eliminates the need to move massive datasets to separate vector databases.With this, you can search and discover the data you need. Applying it immediate ready for use to any AI agent and chatbot.

What this does is creating a Global Metadata Engine across all ONTAP Data Services whether on premises, in the cloud or even as a software running as a VM in virtual environment. Any consumption model of choice. The first in the world!

Check out ÅFX performance here.

2. NetApp Ransomware Resilience Service (link)

In the past, customer would need to stand up their own clean environment for Ransomware recovery and forensic purpose. With the new service, NetApp Ransomware Resilience Service is able to provide an isolated and clean for such purpose. This definitely allows customers for faster recovery and lower the chance of repeat attack.

3. NetApp Cloud Volumes Enhancement (link)

NetApp is the first and remain the only to provide first party cloud data services. Together with the 3 CSPs, NetApp together with Azure and Google, provide further enhancement to this services.

4. NetApp Workload Factory (link 1, link 2, link 3)

One key area of improvement is reducing the SSD capacity. Customer can now increase or decrease the usage as needed. Additional of support for Oracle deployment and SQL backup via SnapCenter integration.

5. NetApp Trident 25.10 (link)

New NetApp Trident 25.10 will not just support the new AFX but also will have long waiting automatic failover of your Pods. In additional, it finally support FC on the ASA block storage. It remains cost free to all NetApp customers and support all major kubernetes distributions.

6. NetApp Shift Toolkit (link)

A free migration tool to move between VMware vSphere and Microsoft Hyper-V. It also support migrating from VMware vSphere to major KVM hypervisor such as RedHat, Proxmox, Oracle Linux VM, etc. Leveraging of NetApp ONTAP FlexClone, it measure to take up as low as 5mins per VM.

7. NetApp Console (link)

Rebranding of NetApp BlueXP to NetApp Console. Making it more seamless integrated into NetApp total offering portfolio. This also introduce an overhaul of the canvas.

Check out more details when it is release.

The Most Secure Storage on The Planet - NetApp

2025-09-18T21:30:00.010+08:00

Why NetApp is bold enough to say they are the Most Secure Storage on the planet second to none.

Here I am going to break it down what NetApp proof itself on:

NetApp has pledge it self to CISA Secure by Design Commitment. And you can read here what that is about also what the pledge entails.
NetApp also comply to NIST Cybersecurity Framework 2.0 by folloing the stringent guidelines entail here.
NetApp has comly to a list security certifications as details here. To list a few, DoDIN APL, FIPS 140, Common Criteria, etc.
Lastly, NetApp is validated by CISA for CSfC to store secret and top-secret National Security Systems (NSS) data.

That is all what NetApp has proof. Why not let a third party do validated.

NetApp engaged SE Labs to test out its built in Autonomous Ransomware Protection (ARP) with AI against 15 ransomware family, and has achieve a AAA rating. The only storage on the planet so far to get this rating. It proof that ARP was able to detect all 100% accuracy of the attack and able to protect 99% of all such attack.
This year 2025, NetApp has also reinforce this status.

Lastly, this is all made possible with NetApp secret sauce, ONTAP, the operating system that sit in its unified solutions. This also means that ARP is available on the on prem hardware appliance, ONTAP Select (Virtual Machine) and Cloud Data Services such as AWS FSxN.

Coming ONTAP 9.17.1, NetApp will also offer ARP to its Block Storage on ONTAP. This will also be offered by cloud offering later once ONTAP 9.17.1 is adopted by the Cloud Service Providers. Having block ransomware protection will be very applicable for database and virtual environment such as VMware, RedHat, etc. This will further provide any storage even HCI such as vSAN or third party storage if it is front by ONTAP Select. Giving customer the ability to centralize manage using NetApp BlueXP.

This makes NetApp the First and ONLY vendor that offers ransomware protection for File and Block.

With all the above, do you think NetApp live up to be the Most Secure Storage on the Planet?

VMware vExpert 2025 Second Half Application is Open!

2025-07-08T00:38:00.005+08:00

For those who missed the first half application or wasn't to the cut and have made improvement as suggested from the comments given, the application for second half is now open!

If you do not know what is VMware vExpert, you can check out my post here. You might be already doing it why not apply and get accredited.

So start drafting and save your application and remember to submit them by 24th July US.

You should expect the results to be out in Aug time frame. As usual if you need help or advice, feel free to approach any of the vExpert Pro nearest to you.

If are not aware of the benefit and experience of vExpert accreditation check out my post here.

Good luck to all applicants.

Dates for reference

Open July 7th
Closing July 24th
Announcement Happens on August 15th

Why NetApp?

2025-06-17T21:30:00.014+08:00

When I join NetApp from VMware after spending a good 13 years, I was amazed and taken aback how much NetApp got to offer to customers. It not just a Data fabric company but lots more in the space of how data are managed, stored, secured, and much more surrounding data.

The first thing I learnt about is ONTAP, the powerful software operating system that powers all NetApp hard and soft boxes at the very beginning. It also not just a storage but it is the King of Files for a reason. The unique of NetApp is in its way of storing data, Write Anyway File Layout (WAFL). Its ability to handle large amount of small files makes it elite on its own. Today NetApp is the ONLY vendor able to handle File, Block and Object all at the same time with ONTAP.

NetApp also have acquired a few solutions over the years specific to how data are created, managed, used and secured. Let's leave that to another post for now.

Let me drill into 3 key takeaways on why NetApp how it differentiate itself as the only one.

1. Most Secure Storage on the Planet

A very bold statement is true in every aspect. The ONLY storage validated by CSfC program and together certified with security certifications such as FIPS140-3, Common Criteria among many others. It is also aligned with the full NIST cybersecurity 2.0 framework.

NetApp is the ONLY vendor with Autonomous Ransomware Protection built in, no third party solution or complex setup with external components and tested by SE Labs with 100% legitimate detection and 99% accuracy in True Value (99% Recall 100% Precision) with a AAA rating. Not any marketing or confidence value that has no backing to the number. With such confident, NetApp also include Ransomware Protection Guarantee.

NetApp is the ONLY storage that include multi-admin verification to ensure every secure task cannot be completed by just one single individual. Imagine just like in a movie launching a missile, you need multiple general approval to activate the system.

NetApp also provide Cyber Vault which provide Immutable (cannot be changed) and indelible (cannot be erased) volume for protection and safe keeping from Cyber attack which can extend from your primary storage to secondary storage. This is a one way "pull" Snapshot. Very suitable for logical air gap.

End to end encryption with data taken from in any form ensuring data at rest and in transit.

Watch how you can leverage on all the functions on NetApp storage and offerings to help you fully protect your data.

2. Data Protection Every Where, Every Way

NetApp storage are bundled with ONTAP One. The software that provide all the feature functions as part of the core operating system for FREE.

NetApp SnapCenter software is a data protection management software that protect enterprise applications, databases, virtual machines and file systems. It is not just a backup for supported critical applications and database providing Application-Consistent Data Protection. Creating multiple copies via FlexClone which does not consume additional capacity. SnapLock allows tamper-proof snapshot against cyber threats or compliance reason which requires a holding period. Also to use NetApp native Snapshot for use for backup, SnapDiff allows to take snapshot from a difference between snapshots instead of going across the file system to check for different. NetApp integration with Backup Vendor by allowing Backup Vendors to utilize SnapDiff framework.

One way of replicating for backup copy purpose via SnapVault. Volume is replicated one way to a destination volume. This is very efficient due to the snapshot technology from NetApp ONTAP with only changed blocks are replicated after the first snapshot. This is to backup to a secondary location providing long-term data retention and recovery options use case. You can check out the best practice guide using SnapVault.

SnapMirror replicate across two volumes and having both consistent and able to fall back to each other for a data restoration. It is useful for disaster recovery and data migration from on premise to the cloud or between ONTAP systems. And move back when there is a need for migration or data restoration. It leverage SnapDiff to identify the snapshot changes.

Targetting protection, NetApp works with backup vendors such as Commvault, Cohesity, CommVault, Veeam, Rubrik, etc. leveraging on NetApp SnapDiff on NetApp storage using NetApp Backup Framework to perform backup quickly than any solutions.

On top of usual applications, containers protection are included via NetApp Trident which is included as part of NetApp system. CSI driver are provide to allow developer to leverage NetApp systems to orchestrate and protect their container pod and namespace by provision and manage storage without the need of storage administrator. NetApp Trident also support and tested with VMware Tanzu as well as RedHat OpehShift.

For a metrocluster solution to keep two active sites managed as one, NetApp MetroCluster provide continuous availability.

Which is a very extensive data protection offered by NetApp.

3. Any Consumption Model Managed as One

Single Management Pane with NetApp BlueXP. NetApp BlueXP is more than a management pane. We will leave that on another day. With one of its capability as a single management pane, you will be able to manage your NetApp systems, virtual machine, cloud storage, all in one place to configure, manage and run. This come as part of any NetApp storage solutions.

Hub and Spoke through FlexCache. Ever have edge or remote office that requires data from Headquarter and requiring to copy all the data and constantly keep up to date will be a challenge. NetApp has the solution for you, NetApp FlexCache. Even if a file is required, only those blocks that are needed are fetch from Hub and not the whole file. This helps customers optimize bandwidth utilization and also optimize egress traffic if you are using any of the NetApp File Services from the CSP as a remote or edge location.

Running any remote or edge location with a KVM (such as RedHat, Proxmox, etc.) or VMware virtual environment with a Enterprise Storage function requirement? NetApp ONTAP Select is a software offering for deploying in a virtual platform as a Storage VM or into vSAN to further extend its capabilities.

Any Data, Any Place:

AWS FSx for NetApp, Azure NetApp Files, and Google NetApp Volumes is the ONLY first party storage running in Cloud Service Providers (CSP). It is managed and purchase directly from CSPs. The same concept and experience in terms replication, data lifecycle, etc. are all the same.

If you prefer to manage your own Storage Service, you can use Cloud Volumes ONTAP. This will provide you NAS and SAN for your applications using NFS, SMB or iSCSI.

If you are running solution such as VMware on CSPs, you can leverage NetApp Files services from our CSP with VMware on X. Amazon Elastic VMware Service with AWS FSx for NetApp, Azure VMware Solution with Azure NetApp Files and Google VMware Cloud with Google Cloud NetApp Volumes.

In summary, application data matters! It doesn't' matter where it sit whether in your data centers or in public cloud, applications data are protected and secured. That is how NetApp provides an Intelligent Data Infrastructure for your data, Here, There, Everywhere!

For any of NetApp features and functions, you can try out lab on demand https://labondemand.netapp.com/.

Is A Shift in Data Management Trend Happening?

2025-05-15T21:30:00.007+08:00

Looking at the trend of data referencing from here. We can see that data has been growing phenomenally. In actual fact, looking at the trend, the boom started in 2020 and has not stop ever since.

Source: https://explodingtopics.com/blog/data-generated-per-day

In today infrastructure, we have two major solutions for data management. One is our traditional storage array and the other is HyperConverge Infrastructure (HCI). The latter only started picking up in early 2011 where some of the known names such as pivot3 and the major players after such as Nutanix and VMware vSAN which exist most commonly in most data center infrastructure.

As data volume increases and the the adoption of AI, capacity and performance for data management become a crucial requirement. Taking example from the like on virtualization, you can have all the compute resources for your virtual machine, but if your workload data IO and capacity cannot be serve, the bottleneck still exist.

In the world of AI, the same thing exist, where you can supply the compute (GPU), but if your IO and Network bandwidth cannot keep up with the demand, the results of training of query in production can be a difference in leap and bounds.

Let's take a look at how some of the storage technology such as HCI has evolved. VMware vSAN has always to run HCI with external storage. It all depending on the type of workload which will fit in each.

While at Nutanix, the "No SAN" message started in 2011 has somehow take a step back recently. Nutanix has announced to work with Dell to support external storage, PowerFlex. The first since its "No SAN" messaging. With the recently announcement to work with Pure Storage is definite an assurance that Nutanix is taking back their no SAN messaging.

Just looking at the market trend especially using HCI adoption and transformation, it seems that external storage is still very relevant and is making a come back after so much consolidation into HCI solution whether is it due to performance or capacity, a 2-tier or 3-tier is still much needed to meet the data growth and demand.

The real move to HCI in the past was all due to cost vs performance and enterprise ready features. The cost of running HCI may no longer run as "cheap" or cost effective. Most did not cost in the compute resources needed to run the storage function or controller VMs, the licenses, the complexity of hardware needed (SSD, HDD, Controller and all the firmware on them and those on the servers) that is needed to work with the HCI software. Catering enough server resources to perform storage activities on top of ensuring performance of the workloads can results in higher cost. Extensive monitoring is often needed. Scaling performance and capacity of storage cannot move away from adding more compute resources which often go wasted.

The cost and complexity has trigger some customers to move towards external storage where data management can be keep in control and not tying with compute resources. Also HCI still lack much of what most standalone external storage provides in many enterprise features, functions.

Next looking at HyperConverged. Taking example from VMware with EMC on VCE, and NetApp, VCN followed Dell with to VxRAIL and VxRack. The investment from customer to start with is huge. The R&D cost from all 2/3 companies was not easy to maintain. To add on top of the overhead, a high cost with high complexity where 3 companies to maintain the hardware stack in terms of hardware firmware, software upgrade and patches and the delay the catch up of bug fixes and security vulnerability.

In my past experience, its takes almost a month for such integrated solution to come out with an integrated solution or release. We can see what has happened to VCE, VxRack and VxRail. Even the adoption of VCN hyperconverged is not that great too. Time can be shorten but I do not see any possible way of shortening due to multiple vendors involvement where priority of R&D effort need to share among each of their own solutions and 3rd party partnership.

Either they have disappear or the adoption has been really poor. The change in EMC by Dell and VMware by Broadcom outcome resulted the change in such integrated solution. Can we ensure such incident would not happen and resulting the full stack to be phase out in the future? It just need one change in one of the vendor in the total stack to make the complexity happens. As a customer, what will be the only option is to migrate everything out to 2/3 tier architecture which will be a massive work all together again.

Some consideration to take into mind is how easy is to move out from commodity hardware as replacement such as moving back to a 2 or 3-tier architecture? Will you as a customer be able to find immediate replacement? How quickly can it be done, weeks, months, years?

What do you think, is external storage still a thing? Will HCI take over the market? Which will be a higher adopter in the end?

NetApp Supports All Virtualization Platforms

2025-04-29T21:30:00.001+08:00

You hear me right. Was browsing through some documentation and wanted to see any storage vendor is supporting Proxmox, the new player in virtualization. To my surprise, NetApp not only support, but all functions of ONTAP are also supported just like how it supports other virtualization platforms such as VMware, RedHat, Microsoft. Any virtualization platforms that support external storage solutions.

Check the virtualization overview documentation site. NetApp has extensive partnership with VMware to support day zero feature function. This can be seen from the support of vSphere vStorage APIs for Array Integration (VAAI) on vVols. RedHat OpenShift, Microsoft Hyper-V and Promox.

NetApp allows customers to choose to consume storage in any shape and form namely, on premies physical storage array, in the cloud as first party files services from AWS, Microsoft Azure and Google Cloud, Cloud Volumes as third party service in CSP's market place and ONTAP Select (virtual appliance).

With true NetApp data solution, you will be able to not just provide performance, availability, efficiency and also security across all offerings. To add on top, even to ensure your critical applications and containers workload be protected with no extra cost!

I am truly amaze what NetApp is doing and helping customers to provide a resilient data platform but also future proofing by staying ahead and giving customers the best return of interest. With 30 years of R&D effort just to provide the best Intelligent Data Infrastructure. Who else beats that?

Every day learning something new!

NetApp - The Intelligent Data Infrastructure

2025-04-17T11:44:00.003+08:00

This week marks my 90 days in NetApp. It was truly an eye opener to what thought to be a storage company can do beyond just data storage. NetApp has evolve over the years and with the amount of R&D and investment made to its portfolio that I believe most are not aware just like myself.

In my 90 days, I should address what NetApp offers in a nutshell:

1. The only true hybrid and multicloud solution that allows customer to consume storage solution in any shape and form with the same skillset and knowledge.

NetApp's secret sauce, ONTAP is the underlying software that empowered its storage solution since the very beginning as the King of NAS. Of course as a software, it has been able to support block and object similarly well. ONTAP One is the only soft license that comes in all NetApp unified storage solution at no additional charges.

But the power of ONTAP does not stop there, ONTAP Select, a virtual machine offering allows customer to consume the same ONTAP storage, right off virtualization platform such as VMware and KVM favours. It can also be found in the major CSP market place. This is often very useful in a remote, edge, IOT or even small projects or offerings that some customer build to offer to their customer. Not to mention, it supercharge HCI solution such as VMware vSAN completing it with enterprise like features. Giving customer the full control of the storage solution right to the versioning of their choice.

Similarly, the very same ONTAP can be found offered as the only first party* offerings by the major CSP namely AWS, Azure and Google since 2013. Where they are offered as native services, sold and managed by the respective CSP.

Note: *first party definition means to manage and sold by the CSP as part of their native service offering.

With multiple data sources, NetApp BlueXP allows customer with a centralize single pane of management to operate them all from a single place. Whether to set up data classification, data protection policy or even replication between sources, BlueXP does it all from one single GUI as what it does best, a Modern data estate operations made simple.

2. When we talk about data management, one key area is on data availability and protection. NetApp SnapCenter, a true native data protection in backing up your data supporting a wide range of databases out of the box is where it shines. Together with NetApp Trident, it even covers modern applications running kubernetes and help in provisioning and protection with ease on all NetApp sources. Top key area of critical capabilities on primary storage platform, NetApp come in top as shown below by Gartner.

Reference: https://www.gartner.com/doc/reprints?id=1-2IUGCBR5&ct=240919&st=sb

On data availability, especially area of Cyber Resilience, Autonomous Ransomware Protection (ARP) is one of NetApp unique feature provided as part of ONTAP. Which makes also NetApp the most secure storage on the planet. As the last line of defence in ransomware protection, NetApp ARP is able to detect and protect from ransomware in seconds (tested by SE labs). There is no solution which is able to do that due to lack of access to data. There are no vendors is able to provide ARP thanks to ONTAP. The secret sauce that makes NetApp unique.

On top of that, NetApp is the only storage data management that provide Ransomware Recovery Guarantee.

3. In the area of data, NetApp has partner with multiple partners on data warehouse and data lakes such as SnowFlakes, Domino Volumes, DremIO, and many more. One area is also in AI, where during GTC 2025, NetApp has gotten the certified Nvidia SuperPod storage on the AFF A90 storage array.

On top of that, ever wonder a simple query of data from multiple sources in the AI space will need performance to return results and data movement will be everywhere. Not to mention, knowing where your data resides, is already a challenge.

NetApp as announced during NetApp Insight 2024 with it's whitepaper, with a disaggregation and composable architecture for future ONTAP and as well as any data be it on prem, cloud, STaaS, any ONTAP storages will be able to benefit from ONTAP Data. In short, a purposeful built solution to be able to contain all metadata at one place where data can be from any location and be able to query those metadata without having to access them physically or transferring them across or even knowing where they reside. You will only need the data when you really need them by accessing it. This will be a very exciting solution where it has been an existing challenge where NetApp has just the solution to address it real soon.

While I continue my journey in NetApp, I am looking forward to NetApp Converge 2025 and see what awaits in the coming month.

NetApp is not just another storage vendor, NetApp is the only Intelligent Data Infrastructure solution priovider.

VMware is making ESXi free for 8.0U3e

2025-04-11T21:27:00.005+08:00

Yes you hear me right. Check out the vSphere 8.0U3e release notes.

VMware is releasing the ESXi 8.0U3e as an entry-level hypervsor.

To download, head to https://support.broadcom.com/ and create an account if you have not done so.

Follow the screenshot below: Software> VMware Cloud Foundation>My Downloads

Next click on Free Software Downloads available HERE.

Scroll down to VMware vSphere Hypervisor and click on it. You will be lead to it.

You will be able to download it.

It will be about 618Mb

That is definitely a good news for those who needs it for Home lab testing. But not sure if this going to be a one off. Also to note, it is on a 60 days trial but not sure if it behaviour the same as pre-broadcom after expiry.

Thanks to Joey Ware for tipping off in Facebook group.

VMware VCF Minimum Cores Purchases Changes

2025-04-10T21:30:00.036+08:00

As per the article and confirmation from email from Broadcom to all their partners, there will be a minimum purchase for New VCF license purchase for new order starting 10th April 2025.

What will be your impact to customers like yourself? Basically for any new project, a minimal purchase of 72 cores license is required whether or not your compute workload is going to be lesser. This does not change the minimum purchase of 16 cores per CPU which still stand.

While that is true, the above statement only applies to new purchase, if you are doing expansion or renewal of licenses, this does not change.

So if you really just need a small setup using 3 servers each, taking a supported Dell PowerEdge R350 with vSphere 8.0U3 as per VMware site, with a minimum of 8 cores per CPU, you will need 24 cores license which will have a surplus of 48 cores (72-24). That is a bumper.

For the above example, this will be wastage. However, as I mentioned, expansion and renewal are not impacted.

In another scenario, if you have multiple remote sites each with the same setup above, instead of treating it as a new project each, you can cluster them as one project, purchase a minimum purchase of 72 cores and use 24 cores of it. Add on servers for other sites as expansion which does not need a minimum and utilize the remaining 24 cores. Moving forward a more proper planning to optimize your license purchase is definitely required instead of loosely buying them in the past.

This will be something to consider which means you will need to plain way ahead when you are just looking at a small project, think if you will be expanding or not to treat it as a one big project for a purchase so that you can minimize your wastage and utilize your license at full.

One thing to note that if you need more capacity of storage using the above example on VCF 24 cores, you are entitled to 24 TiB of capacity. If you need more licenses for VMware vSAN, each vSAN add-on license will set you an additional $210 per TiB.

The question will be, if you need more capacity, does it provide enough performance and data management with protection capability? Is vSAN sufficient to fulfil those requirements with the cost? If not, how much more other solutions add on. Or can it be really be met?

You might want to consider moving to 2-tier architecture by using an external storage such as NetApp ASA. This not only allows your to separate out your management domain supported in VCF 9.0 to avoid operation overheads and have availability outside of the virtual platform.

NetApp ASA is designed for block protocol and suitable for VMware environment which not only provide the storage requirement in capacity but also performance and data management with protection, running the renowed ONTAP.

Did I also mention performance and data availability of six 9's guaranteed? No drivers installation and getting full symmetric active-active path which only found in high end storage. Trust me, I couldn't believe it myself untill I am in NetApp. Dont't let me tell you, go check it out yourself.

Supercharge VMware vSAN with NetApp ONTAP Select

2025-03-18T21:30:00.026+08:00

VMware vSAN is a Hyper-Converge Infrastructure solution which is part of VMware Cloud Foundation (VCF) or VMware vSphere Foundation (VVF) bundling.

While VMware vSAN helps customer to achieve a SAN like storage pooling using local HDD on each server together, it is still limited on features it encompasses in comparison to a physical SAN.

This is where NetApp ONTAP Select which will complement the missing gap. ONTAP Select is the virtual appliance of the well known ONTAP that runs on all NetApp Unified Storage appliance.

One of the major function that ONTAP Select brings to vSAN is on Data management and Protection. This is very useful for protection specific applications especially application aware application by ONTAP Select. Some of the enterprise applications supported can be found here. A detailed kb from Broadcom can be found here.

With ONTAP Select, these applications that are supported can be protected via SnapMirror which can provide an application consistent snapshot (vSphere Snapshot is not application aware) or replication across two different cluster to another ONTAP Select cluster or ONTAP Appliance (Storage Appliance) vis SnapMirror. This gives customer not only data protection but also provide flexible option in planning for data management and disaster recovery options. Such as from edge/remote location running ONTAP Select to a main site running ONTAP Appliance, etc. Or running ONTAP on site replicating data to ONTAP file or Cloud Volume in the cloud such as AWS, Azure and GCP. SnapMirror also support with VMware Site Recovery Manager (SRM).

With the ability to deploy ONTAP Select onto VMware vSAN, this increases the capability of VMware vSAN, and providing customer with some of the ONTAP features that are not available on HCI solutions.

Lastly, ONTAP integrate with VMware vCenter to help in managing the storage more efficiently. That's what we call better together.

ONTAP Select can run on KVM to provide storage protocol such as NFS and Block to any systems.

Check out some of the resources below on ONTAP:

Requirements for ONTAP Select on VMware vSphere
System Requirements for ONTAP Select
Supported features for ONTAP Select on VMware vSphere
ONTAP Select licensing
ONTAP interop
ONTAP tools for VMware vSphere

VMware vExpert 2025 VMUG Advantage

2025-03-17T22:37:00.001+08:00

First of all congrats to all vExpert 2025 announced. This year, there is some additional entitlement not made known to the community. I was surprised when I received the below email:

And yes, you got it. All vExpert 2025 are given a year of VMUG Advantage entitlement. This includes a 50% one time certification fee and many more which you can find out more here.

This is definitely an icing on the cake.

The New Era of Data

2025-02-18T21:30:00.087+08:00

Back in the days of VMware, I was fortunate to talk to many customers and involved in numerous discussion on helping them from on premise to private cloud, hybrid cloud to public cloud and MultiCloud to today, Intelligence running in the cloud.

Many thought that might change when I move on to my next career. In fact, I carry on that discussion with my customers today the same way I did in the past. That was also one great reason I joint NetApp.

Let me show you where this happens:

Reference link

As changes of IT evolved, the amount of data customer have to collect, manage, protect and secure is way more than before. More so around complexity and managing the amount of data threats.

In 1996, NetApp introduced ONTAP to solve the issue with local data siloes with NAS to help customer manage the large amount of data in files and block.

When virtualization take place, NetApp supported a wide range of solutions such as VMware, Hyper-V, KVM, etc. helping customers manage that data growth. Even so adopting all the VMware vSphere Storage VAAI function such as vVOLs. Supporting both on private and public cloud. Which is termed, Hybrid Cloud.

Next came the Multi-Cloud era, where the cloud service providers (CSP) such as AWS, Microsoft Azure and Google Cloud came. NetApp carrys the same strategy and became the first/only storage vendor providing first party storage for customer to scale their requirement with their data leveraging the very same ONTAP they know.

Today with AI, Intelligence are baked into. AI Workload demands sky rocketed due to the support requirements from new GPU technology where Nvidia has introduced to the market. What AI has created is performance and capacity growth, this resulted in many organization moving towards a 3-tier architecture due to constraint from other Storage solutions which was unable to cater to the demand.

At the same time, NetApp has announced last year during NetApp Insight to address the growth and performance demand from AI workloads from customers.

The amount of performance and capacity have grow real fast and NetApp have it figure out.

Edited to illustrate Cloud Siloes

With the change in the technology era, NetApp was able to UNIFIED all the Cloud Siloes illustrate above, with a UNIFIED Data management whether it is on premies, virtualized, in the cloud or the drive of AI demand. The same way how NetApp solve Data Siloes, The Intelligent Data Infrastructure.

All thanks to ONTAP. A true software defined storage solution right at the start, that power almost every NetApp storages. Who would have thought and starting a software defined storage back in 1992 when it was first invented?

Doesn't that sound so familiar? Managing the every siloes like how the likes of VMware with their software, NetApp was able to do this right at the beginning on the data layer. Not to mention some of the key advantages such as:

same skillset
same performance
same efficiency guaranteed
same protection
same first party on all public Clouds (AWS, Azure, GCP)

Not forgetting a #1 most secured storage on the planet! Don't just hear from me, check out SE Lab testing giving a 99% accuracy (True Value). 0% of false alarms. Who beat that at your last line of defence when you most need it. With that confidence, NetApp is able to offer Ransomware Guarantee to their customers.

Reference link

With NetApp its all about:

Any Data, Any Place, Unified Control

Reference link

VMUG Singapore 1st Event 2025, Join us!

2024-12-09T09:30:00.005+08:00

The year of 2024 is coming to an end. VMUG Singapore had hosted few events this year and it has been fruitful. They are not stopping there.

Starting 2025, VMUG Singapore is hosting their first event in 2025 on 14th Jan. Do register to join us.

This year I am helping out for this event as a former VMware employee. This time round for a start we are doing something very different.

1. We will still have our product update from Sanjan and Iwan who will be at the first item to answer questions regarding where VMware product is heading and why on what they are doing.

2. We will have our education to update on training and certification changes.

3. For the very first time. We will be sharing career development with speaker, Jeremy Cheong from Juel Search. Some market statistic and giving you some data points on how you can grow your career. We are not just going focusing on technology but also helping our community to grow with us. I will be sharing some job trade secret which companies employee do not share. Want to know more, join us.

Come join us and bring along all your question whether is it career related, solution related, we welcome everything questions you have in mind.

See you there!

VMware vExpert 2025 Applications Starts Now!

2024-12-06T11:12:00.006+08:00

It the time of the year to apply and renew your vExpert for another year for your effort contribution back to the VMware community over the last 6-12months.

The vExpert 2025 application is now open starting from 6th Dec 2024 and will closed on 10th January 2025. So hurry start your application and fill up all your contribution. You do not have to do it once through, you can always save it in draft and add on and submit before the closing date. If you have doubt reach out or look for your vExpert Pro nearest to you to to advise you accordingly.

This year has been a very different year for VMware. However, the vExpert program was valued and carry on. The perks this year been a vExpert will be very different from the investment from Broadcom.

Head over and sign up now. Good luck!

Update 11th Jan 2025

Application dateline has been extended to 24th Jan 2025.

Update 13 Dec 2024

Check out the vExpert application announcement here.

Plain Virtualization

Storage Efficiency • WAFL Internals

Dedupe, Compress, Compact: How WAFL Squeezes Every Byte Out of Capacity Flash

A Three-Stage Pipeline, Not Three Separate Features

The 4KB Problem Compaction Was Built to Solve

Efficiency That Runs When the CPU Isn’t Busy

Why This Is the Whole Point of Capacity Flash

The Quantum Threat Is Real. Is Your Data Already Protected?

The Quantum Threat Is Real. Is Your Data Already Protected?

The Reality of Infrastructure-Wide Security

The Quantum Wake-Up Call — and Who Was Already Ready

A Security Pedigree Built Over Two Decades

Common Criteria: Two Decades of Independent Validation

NSA CSfC: Trusted to Store Top Secret Data

CISA Secure by Design Pledge

Ransomware Protection: Built In, Not Bolted On

The "Last Line of Defence" Principle in Practice

The Question You Need to Ask Today

References & Sources

Found this useful? Share it.

Harvest Speaks: Turning ONTAP Metrics Into Conversations

Harvest Speaks: Turning ONTAP Metrics Into Conversations

What Harvest Actually Is

The Familiar Architecture

The Pivot: From Dashboards to Dialogue

Leveraging an Existing AI Agent

What This Looks Like in Practice

Why This Matters Beyond the Ops Desk

A Few Things Worth Keeping in Mind

Closing Thought

Your AI is Ready. Your GPU Isn't. Here's the Fix.

The GPU queue problem nobody talks about

The data gravity trap

What is NetApp FlexCache — and why is it different?

The compliance advantage: just enough, for just long enough

Where this matters most in enterprise AI

Where FlexCache fits in a broader AI data strategy

You only pay for what you actually use

Going further: when an AI agent manages the burst itself

Further FlexCache capabilities worth knowing

The question is no longer whether to burst — it is how

Garbage in, AI out. Why 60% of AI projects are set to fail.

Garbage in, AI out.Why 60% of AI projects are set to fail.

AI adoption is accelerating — faster than any technology before it

The uncomfortable truth: data is what’s missing

Data: the heart of everything

What good AI data engineering looks like

Three things to get right for AI success

The ecosystem that completes the picture

Beyond the Specs: Pre-Sales Architect Value in the Era of Flash

The "Storm" Before the Calm: Managing the Boot Storm

Policy Optimizations: Trimming the Fat

Antivirus & Backups: Don't Let Security Slow You Down

Key Takeaways for the Modern IT Leader

Final Thought:

Storage in a Seller’s Market: Flash Prices, Licensing Shifts, and Why IT Leaders Must Move Faster

The VMware Effect: When Licensing Strategy Becomes a Market Signal

Why Customers Felt Cornered

Vendor strategy can change faster than your ability to migrate.

NAND Flash Shortage: Déjà Vu With Higher Stakes

What’s Happening on the Ground

Vendor Responses: Different Speeds, Same Direction

Rethinking Hyperconverged Infrastructure in a High-Cost Era

Hybrid Strategies Beyond Tering: Cloud as a Pressure Value

The Illusion of “Better Deals” in a Shortage

Option 1: New Supply Chains (e.g., Chinese Vendors)

Option 2: Vendors Holding Prices

Flash-Only Architectures: The New Lock-In Risk

Rethinking Optimization: Hybrid Is Back

Cost-Control Strategies Regaining Relevance

Trade-offs to Acknowledge

Procurement Must Evolve: Speed Is Now a Competitive Advantage

The Old Model

The New Reality

Key Considerations to Avoid Future Lock-In

1. Architectural Flexibility

2. Supply Chain Resilience

3. Licensing and Commercial Model

4. Scaling Economics

5. Procurement Agility

Garbage in, AI out.
Why 60% of AI projects are set to fail.