Data Center Strategies

Microsoft’s licensing revisions and its effect on disaster recovery

2008-08-21T06:37:25-07:00

Chris Wolf and I were briefed by Microsoft on their licensing revisions for their server based applications that make the licensing terms more favorable to virtualized environments. Chris blogged on these details here. There’s another aspect of the licensing revisions that I want to talk about: Disaster Recovery in your Business Continuity plans.

Prior to the revisions, a user could not transfer an application license to another physical server more often than once every 90 days. Legally, this didn’t allow for disaster recovery testing with only one license for your application instance, nor for any type of disaster event that would result in failing a service over to a recovery site for less than 90 days. You would need to either stay at your failover site for 90 days minimum, or you would need to purchase additional licenses for your recovery sites, even though they would not be in use except during a disaster event or testing.

The new licensing terms are more favorable to disaster recovery and business continuity practices, but not for everyone. The new terms define the concept of a “server farm” which is composed of no more than two data centers that are no further than four time zones apart from each other. You can transfer a license between physical machines as part of virtual mobility, disaster recovery, or business continuity within your server farm as often as you like.

My first reaction to the "server farm" restrictions were "what?!?" and "why would you do that?". So I questioned Microsoft during the briefing as to why they chose to define a “server farm” as no more than two data centers located within four time zones of each other. Their answer was clear: They do not want customers transferring a software license around the world within a 24 hour period as in a “follow the sun” fashion. They regard this as misuse of a license. Personally, I see this restriction as unnecessary. As long distance data transmission latencies decrease, the business division of a world wide company in Japan can use the same license as the division in New York just by both remotely accessing a common data center in Colorado. Why make the restriction in the first place? I know of a smaller company located in the Phoenix, AZ area that placed and now remotely manages their production servers in a co-location center on the Isle-of-Man (British Isles) to save money.

So what does this mean for you the customer? It has a series of implications. For those who have disaster recovery centers within four time zones of your production data center, you now only need one license per covered Microsoft server application instance. This can translate into savings in your business continuity plan. However, those of you who have off-shored – or are looking to off-shore – your disaster recovery solution, you may not be able to reap the benefits of this licensing revision. You will want to take this into account when calculating your potential savings by off-shoring – it may change your plans. Those of you who have already setup disaster recovery data centers of your own across an ocean, such as a primary data center in New York City with a recovery center in Ireland, will not be able to take advantage of these licensing revisions between those data centers.

Note that Microsoft specifically indicated that they want to discourage license movement to “follow the sun”, as you can reap the benefits of the licensing revision in a long distance recovery site if you setup a data center in North America with a disaster recovery center in South America, for example.

[Posted by Richard Jones]

SSD love - its in the air

2008-08-19T12:33:58-07:00

Coming off the recent Flash Summit I have to admit that I'm duely impressed with the progress in the solid state disk arena. Love (or should I say hype) is definitely in the air for these devices. To hear the many speakers at the Flash Summit (held in Santa Clara) espouse the wonders of SSDs, one would think that hard disk drives are the nasty step child better forgotten. The Summit was well attended - in one case there were so many folks attending a SSD design session that the session was moved to a bigger room because the hallways were blocked with the overflow of SSD enthusiasts. I like it - excitement in the storage biz is always a good thing! Who says that storage is "borage"? Not anymore! But there's hype, hype, hype...lots of hype!

Everything was going great at the Summit, forum speaker after speaker singing praises for SSDs until near the end of that morning's session, a speaker from Fujitsu (I'll withhold his name lest he gets branded a party pooper) turned a garden hose on the participants. His point was clear: where are the buyers of these SSDs? Where are the forecasts by the tier 1 vendors to support the excitement? He wasn't seeing it. In fact, he made the point through projected sales mix of SSDs, 10K, 15K and 7.2kRPM HDDs that SSD sales didn't get interesting for another 5 years - or longer. Door prize to Fujitsu.

So, back to earth, just where are we going with SSDs? Are they in your immediate future? Well, yes and no. And what should you, as the data center CIO/storage administrator/guru, be doing now?

First things first. Learn about the SSD technology! This is not hard disk drive technology. While many of the concepts are similar, there are nuances that are critical to understand. Just so happens I'm writing a document for Burton Group clients to help with this, but for those not subscribed, search the web. There are many resources out there that can help enlighten you.

This is a nascent industry just crawling out of the proverbial swamp. Many vendors will come and go - be prepared for vendor churn. Not every vendor will survive. By some counts there are some 70 vendors in this space. Having many vendors can be a good thing - spurning innovation, leaving no rock unturned - but in the end only the most fit will survive carrying their prized intellectual property (IP) with them. As that IP gets recognized and bartered, there will be vendor consolidation and ultimately the game will shift to the "big boys" and a few lucky upstarts.

In comes Intel. Intel's much anticipated announcement today is just what the SSD industry needed. A power house company endorsing these new fangled SSDs. Even better is that the product line spans from laptops to enterprise applications. This is an important step forward, bolstered by support by Sun (big), Lenovo (bigger) and HP (biggest). Good news no doubt for customers of laptops and soon, even though I know EMC has been doing this for a while, enterprise storage.

Ok, before we get too excited again, take note: pricing for SSDs must come down and soon for the laptop market. Yes, many a marketing type likes to espouse the price of SSDs based on IOs compared to HDDs - a handsome comparison - but that's not what consumers by and large will understand. Does your mother understand IO transactions per second? I doubt it. She wants more for less. She wants a snappy responding laptop. SSDs for laptops need to be cheaper - and noticeably faster - for wide scale acceptance. Consumers will lead the way.

Back in the enterprise, the place where you can lose your job for jumping on a new technology too soon, things are a little more cautious. Yes, data centers are hungry for more performance from their storage but not at too high a risk. And no doubt the risk seekers will be early adopters of SSDs. But the bulk of enterprise customers (aka CIOs), I'll predict, will be watching and waiting for the credibility of the new technology to build. A key inflection point will be when they realize that their own laptops contain an SSD, and then the love will flow.

Now that Intel has arrived with credible products, not to say others don't also have decent SSD products, those others just lack the financial muscle. Watch for other big players like Samsung and Seagate to show up at the enterprise class data center SSD party. Then it's Game On: enterprise storage subsystem vendors will be all over themselves designing SSDs into their boxes looking for the love. And then maybe Fujitsu, best wishes always, will be proved wrong and be late to the party.

Posted by Gene Ruth

Interpreting Microsoft's Revised Virtualization Licensing Policy

2008-08-19T06:21:50-07:00

Today Microsoft announced major changes to how several Microsoft applications (e.g. Exchange, SQL, SharePoint) are now licensed for virtual environments. The most significant part of the change is the removal of the 90 day license transfer restriction that had previously negatively impacted Microsoft application licensing costs in virtualized environments. All Microsoft licenses are assigned to physical servers, so if a VM is moved to a server without an available license for a given Microsoft OS or application, Microsoft considers the VM movement a license transfer. Once a license is transferred, it cannot be reassigned to another physical host for 90 days. This policy has long impacted virtualization high availability and live migration.

To understand the problem, consider having four VMs running SQL Server 2005 in an eight-node VMware ESX cluster. Per the previous policy, you would need 32 (8 X 4) SQL Server 2005 Standard Edition licenses in order to remain in compliance, given that all four SQL Server 2005 VMs could potentially be on the same physical host at the same time. Alternatively, you could purchase eight SQL Server 2005 Enterprise Edition licenses, which allow an unlimited number of VMs to run SQL Server 2005 for each licensed physical host. The bottom line - as soon as you virtualize SQL Server 2005, you would have to pay more money to add or upgrade existing SQL Server licenses.

With today's announced changes, you'll now only need to license the four SQL Server VM instances, and licenses could be transferred without restriction across all nodes in the cluster. The financial impact of this announcement to most organizations is probably negligible. Many IT shops simply ignored the 90 day license transfer restriction and had never taken the step to purchase additional licenses for the sole sake of virtualizing a Microsoft application.

It's important to note, however, that the 90 license transfer restriction has only been lifted on server applications licensed under a volume license agreement. Small and medium businesses (SMBs) that have individual Exchange or SQL Server licenses purchased through a VAR (Microsoft categorizes such licenses as Full Packaged Product [FPP] or OEM/System Builder) are not covered under the new policy. So let's apply the 90 day transfer restriction to FPP editions of SQL Server 2005. Assume that a small office is running two SQL Server 2005 systems and plans to virtualize them on a two-node Hyper-V cluster. With today's licensing, the office would have to double their SQL Server 2005 license investment. Since it's possible that the SQL Server 2005 VMs could both be on one server at a given time, you would need two SQL Server 2005 licenses for each physical host. So to run two instances of SQL 2005, you'd need to have four licenses. If you wanted to run them on a three-node physical host cluster, you would need six licenses. At that point, you'd qualify for a volume licensing discount. Volume licensing pricing is available for Microsoft server applications sold in quantities of five or higher.

It should also be noted that the new policy does not apply to server operating systems. So let's go back to the previous small office example. Suppose the office has six Windows Server 2003 systems, each with an FPP license. If those instances were deployed as VMs in a two-node physical cluster, you would need a total of twelve Windows Server 2003 Standard Edition licenses. So again, you have to double your software licensing investment to run the OS as a virtual machine. Alternatively, Microsoft is encouraging organizations to purchase Data Center editions of their server OS licenses to cover virtualization. One server OS data center edition license on a physical host allows an organization to run an unlimited number of virtual machines on that host. So the branch office in my example could purchase two Windows Server 2008 Data Center edition licenses (which include Windows Server 2003 downgrade rights) to cover the added cost of licensing. Of course, if you planned to run Hyper-V as your hypervisor, this wouldn't be a big deal. After all, you need to purchase Windows Server 2008 in order to get Hyper-V anyway. However, if you planned to run any other hypervisor, this clearly represents a significant (and in my opinion unnecessary) cost. This licensing strategy clearly places Microsoft's competitors at a disadvantage, as any other virtualization platform will always be more costly than Hyper-V.

In terms of platform support, you would be hard pressed to find an application or OS vendor that officially supports more virtualization platforms than Microsoft (i.e. Citrix, Microsoft, Novell, Sun, Virtual Iron, and VMware). Also, today Microsoft will announce that VMware has joined the Server Virtualization Validation Program (SVVP) and is now officially supported (see my blog post on the subject).

Microsoft - great move today. I think many enterprises will appreciate the application licensing flexibility that your policy change has provided. Still, let's not forget about the small IT shops that do not have volume licenses. Many of them are likely candidates to run Hyper-V, so why make them double their Exchange or SQL Server licensing investment, for example, in order to leverage the benefits of virtualization? You made a great step with Vista Enterprise Centralized Desktop (VECD) licensing and allow organizations to deploy virtual desktop operating systems without mobility restrictions. So why restrict server OS mobility? Lifting the 90 day licensing transfer restriction across all product lines is simply the right thing to do. If there truly is a technical reason why OS or application mobility within a data center should be restricted on any level (as it does today with server operating systems and applications not covered by a volume license), please tell me what it is. I welcome your comments.

Posted by: Chris Wolf

It's Official - Microsoft to Support VMware

2008-08-19T05:40:49-07:00

Yesterday evening Richard Jones and I received word from Microsoft that VMware had joined the Server Virtualization Validation Program (SVVP) and I expect an official announcement to come later today.

VMware has now signed on to the SVVP program. Microsoft is certainly excited to add VMware to the program, recognizing the value this provides to Microsoft customers. - Microsoft

This is a really big deal and both vendors should be congratulated on the work it took to make this happen. Once the news is made official, Microsoft applications and operating systems will be fully supported on VMware-based virtualization environments.

Other vendors should take note of Microsoft's support model, as Microsoft supports more virtualization platforms than any other vendor by a hefty margin. Here are the platforms now supported by Microsoft:

Citrix (XenServer)
Microsoft (Virtual Server 2005, Hyper-V)
Novell (Xen)
Sun (xVM Server)
Virtual Iron
VMware (ESX Server, ESXi)

Microsoft and VMware had been working diligently for several months on the completion of their support agreement and VMware's inclusion in the SVVP. We're just weeks away from Microsoft's Virtualization launch event and I'm expecting more good news to come out of Redmond at the launch event as well. Microsoft's official support of virtualization platforms from six different vendors should further accelerate virtualization adoptions throughout 2008 and into 2009.

Posted by: Chris Wolf

Red Hat's Virtualization Moves

2008-08-11T12:16:46-07:00

In mid-June Red Hat became the latest (or should I say “late”) vendor to announce a stand-alone hypervisor for x86 server virtualization based on the Kernel Virtual Machine (KVM). This is a move away from their Xen based virtualization solution integrated into Red Hat Enterprise Linux (RHEL) 5.

The announcement of a beta version of the KVM hypervisor was made at the Red Hat Summit in Boston, MA. While at the summit, I attended the analyst love fests including the Virtualization Panel. It is what I did not hear at the Panel that I wish to discuss in this post. Red Hat introduced the KVM hypervisor as the next phase in server virtualization with Red Hat leading this “new wave”. To me, this immediately raised the question about Red Hat’s commitment and attention to Xen in RHEL 5. So I asked Red Hat if they were going to move to KVM from Xen in RHEL 6. Red Hat beat around the bush, avoiding the question by talking about how great LibVirt (an open source virtualization hypervisor abstraction library) is in isolating customers from hypervisor changes. A simple “yes” or “no” would have sufficed. I drew my own conclusions that Yes they will move to KVM in RHEL 6 (Note: to my knowledge, Red Hat has not made any such announcements about RHEL 6).

In Red Hat’s defense, this would make sense for them to do. They are behind in the server virtualization world. Case in point, they finally released their paravirtualized drivers for RHEL in May, nearly a year behind their competition in this space. With their limited resources, placing their bets on the hypervisor that has been accepted into the mainline kernel will leverage a larger community of open source engineers behind them. Red Hat did mention a few times at the panel that hardware validation and driver support with KVM will simply come in the mainline kernel releases from here on out, and this will benefit customers (and Red Hat more so).

So here’s my opinion: Red Hat should have shown bold leadership in that panel and said RHEL 6 will use KVM as the server virtualization hypervisor from Red Hat. That would have made the direction clear. As it stands now, the impression the audience received was that Red Hat’s virtualization future is not only lagging the industry but is muddy and fragmented.

[Posted by Richard Jones]

Oracle and the Big Elephant

2008-08-08T12:37:35-07:00

The folks at Oracle have done a pretty good job of trying to not pay attention to the big elephant in the room. If you haven't see the elephant Oracle's been trying to ignore, here it is.

For as long as I've been with Burton Group, I have been listening to clients passionately complain about their attempts to get Oracle to officially support running their products on VMware environments. In fact, some have even told of their trials with negotiating "best effort" support with Oracle, knowing that such negotiations occur on a case-by-case basis. Following Oracle's recent announcement regarding the availability of Oracle Virtual Machine (OVM) templates, I felt compelled to ask Oracle my big elephant question - "Will you provide templates for non-Oracle virtualization platforms, namely VMware ESX?" The answer I received was "No." When I inquired about support for Open Virtualization Format (OVF), thinking that Oracle would leverage OVF to distribute Oracle application virtual machine appliances that could run on a number of hypervisors, I again received the answer I wasn't hoping for, "No." Oracle's OVF plans include using OVF to import VMs to their OVM platform, but have no plans to package their own applications using OVF to run on other virtualization platforms. So if you are looking for Oracle support on any x86 virtualization platform other than OVM, you will need to keep waiting.

Now let's turn this around and replace Oracle with another vendor. Suppose Microsoft refused to support their applications on any x86 virtualization platform except Hyper-V? Microsoft today provides "best effort" support for VMware platforms and officially supports Citrix-, Novell-, Sun-, and Virtual Iron-based virtualization platforms. Refusal to support other x86 virtualization platforms is clearly an anti-competitive move on Oracle's part and one that places an unnecessary burden on its customers.Today Oracle is giving their customers two choices: run Oracle applications on their preferred hypervisor and have an unsupported configuration, or split their virtual infrastructure at least two ways and use OVM exclusively for virtualizing Oracle applications. Of course, some large customers have the clout to negotiate support as part of an Oracle software license renewal, but that should not be necessary. Personally, I see such strong arm tactics as doing more harm than good for a vendor.

Oracle - it's time to offer "best effort" support for all virtualization platforms. Practically all enterprise vendors do this today; the vendors will help organizations troubleshoot application faults, but faults related to performance require a virtual to physical (V2P) migration. A small conciliation could go a long way.

Posted by: Chris Wolf

Apples, Oranges, and Hypervisor Price Comparisons

2008-08-07T09:20:52-07:00

Fruit: Apples, Oranges

Fruitful: Current discussions on hypervisor price differences

Fruitless: Getting hypervisor vendors to agree on a fair pricing comparison

The Task

There's been some interesting hypervisor pricing comparisons circulating the Internet over the past two weeks that have caught my interest:

VMware's Mike Dipetrillo: ESX 3.5i for Free and the Impact on Hyper-V and the SMB
Virtualization.info - Alessandro Perilli: VMware ESXi vs Microsoft Hyper-V: which one is better for SMBs?
EvolveTechnologies - Dave Sobel: Is VMware Cheaper than Microsoft?

After reading with interest, I set out to conduct my own pricing evaluation, thinking that it couldn't be too hard to get vendors to agree on a fair pricing comparison (insert joke here). To keep it simple, I decided to use a small branch office consisting of six Windows Server 2003 servers in the evaluation. I consider high availability a requirement in all production virtualization deployments and added virtual infrastructure management to the mix as well. To summarize, the solution requirements consisted of:

Virtualizing six Windows Server 2003 servers on two 2-way physical hosts
High availability failover support
Centralized management of the virtual infrastructure

With the criteria set, I moved to evaluating vendor solutions with each vendor's bottom line list price.

The Proposed Solutions

Table 1 below compares what I see as comparable solutions from VMware, Microsoft, Virtual Iron, and Citrix.

Vendor	Hypervisor Package	High Availability	Management	Total Price
Citrix (Stratus OEM)	XenServer 4.1 (included in Stratus Avance)	Stratus Avance (includes XenServer hypervisor) $2,495 per node x 2 nodes = $4,990	Included with Stratus Avance software	$4,990
Microsoft	Hyper-V (included with Windows Server 2008 Enterprise license) $3,999 per node x 2 nodes = $7,998	Included with the Windows Server 2008 Enterprise Edition license	Essential management is provided by the Hyper-V Manager MMC (included with the OS). Advanced management included in System Center Virtual Machine Manger 2008. VMM 2007 Workgroup edition is priced at $499.	$7,998 or $8,497 (assumed price once VMM 2008 is available)
Virtual Iron	Virtual Iron Extended Enterprise Edition - $799 per socket x 4 sockets - $3,196	Included in Virtual Iron Extended Enterprise Edition	Included in Virtual Iron Extended Enterprise Edition	$3,196
VMware	VI Standard High Availability Acceleration Kit for 4 processors $7,254	Included in VMware Standard High Availability Acceleration Kit	Virtual Center Foundation Server license included in acceleration kit	$7,524

Table 1: Hypervisor price comparison (assumes Windows guest OS licensing is not required)

Citrix XenServer 4.1 does not include high availability, so the Status Avance OEM was used instead. Note that the pricing in Table 1 assumes that the branch office already has existing Windows Server 2003 licenses. If licensing is required, or if plans to upgrade branch office systems to Windows Server 2008 are in place, then the pricing comparison changes substantially. The Microsoft solution already includes licensing for 8 VMs, since Hyper-V is a part of the Windows Server 2008 OS. For the three other solutions, the cheapest way to go would be to purchase six Windows Server 2008 standard edition licenses for $995 each (total cost = $5,970 for six VMs). Downgrade rights would allow you to run Windows Server 2003 under the Windows Server 2008 license, and you would already be covered when it comes time to upgrade. Note that there's no price difference with Windows Server 2008 and Windows Server 2003 Standard Edition licenses.

Table 2 shows vendor solution pricing with the Windows Server 2008 license included (added cost of $5,970).

Vendor	Total Price
Citrix (Stratus OEM)	$4,990 + $5,970 = $10,960
Microsoft	$7,998 or $8,497 (with VMM 2008 once it's available)
Virtual Iron	$3,196 + $5,970 = $9,166
VMware	$7,524 + $5,970 = $13,494

Table 2: Hypervisor price comparison (assumes Windows guest OS licensing is required)

As you can see, OS licensing cost is a significant factor. With the cost of OS licensing, the Microsoft Hyper-V solution changes from the most expensive to the least expensive. Also, it should be noted that both Table 1 and Table 2 reflect vendor list pricing. Since vendor discounts vary, the resulting paid price for any of the above solutions could be thousands less. Still, the list pricing comparison should give you a good idea of where each vendor stands.

The Fruit of My Labor

I had to use a minimal configuration in order to make the vendor comparisons as close as possible. VMware eloquently stated that for an apples-to-apples comparison with Hyper-V, I should include Systems Center Operations Manger and System Center Configuration Manager in the Microsoft price. However, while Virtual Center does a few things that Virtual Machine Manager doesn't do, System Center Configuration Manger and Operations Manager can do much more than Virtual Center. So to balance the comparison, I would have to add System Center Configuration Manger and Operations Manager to the price of each solution, making the addition a wash. The same happened when backup was added to the equation. Citrix and Stratus would also argue that their solution includes local storage mirroring support, so purchasing shared network storage is not required and could represent considerable savings. VMware also noted that their price includes a year of Gold Support, and that similar support on competitive offerings could raise the price as much as 20%. Since I cannot buy a VMware product without support, I'm keeping the support cost in my comparison because it does represent the minimum price. In addition, Virtual Iron's competitors have been quick to note that Virtual Iron does not have a similar supporting vendor ecosystem, which is a decision making factor you won't find on a product data sheet.

So I'm sure vendors will still argue along the lines of apples-to-apples, but I counter that I am comparing apples-to-apples while agreeing that some may be tastier than others. You may see some of the solutions as granny smith apples and others as red delicious apples. On the outside, they're all apples, but on the inside the experience is unique to each one.

Fruitless? Fruitful? I welcome your feedback.

Posted by: Chris Wolf

Storage subsystems include the kitchen sink

2008-08-07T07:46:09-07:00

Your father’s simple RAID array is fast receding in the rear view mirror. Just as cars now offer upgrades such as AC or power windows as the norm, modern storage subsystem are doing the same, throwing in everything but the kitchen sink and setting a new standard of what to expect in terms of features. Sure, some vendors still insist on charging for options but I’ll predict that we are moving to a new place where certain features will become expected – like AC on a car. We are not there yet but I can see the light in the proverbial tunnel – its close and I don’t think it’s an oncoming train.

EMC’s announcement of the new CLARiiON CX4 line is a good example of a full featured array subsystem – EMC may even have included a sink under the covers somewhere. Not to sound like a shill, but evolving, or maybe more appropriately sprouting, from the CLARiiON CX3 line are new features such as:

· Flexible multiprotocol support for fibre channel and iSCSI (nice to have the choice in the same box)

· Enhancements to controller technology – multi core 64bit – for improved throughput and scaling (can never have too much horsepower)

· Efficiency features such as drive spin down and low power SATA (the jury is out as to whether tiering support will make it easy to take advantage of the power savings.)

· Improved replication support (but there’s a license to buy to make it work…)

· Virtual (aka, thin) provisioning ( argh! this “virtualization” thing is getting out of hand!)

And once again solid state disks - SSDs. These interesting devices are starting to show up more and more and now squarely in the SMB space.

For EMC customers, selecting between the Symmetrix and CLARiiON line will be more difficult – likely coming down to scalability and pricing of hardware and support. In many ways EMC, and I suspect the other big boys in the storage business, are following the lead of folks like Compellent, Equalogic (now Dell) and Pillar Data to name a few. All these upstarts have found an eager customer set hungry for more than a run-of-the-mill RAID array. Advanced features such as thin provisioning, replication and simple configuration have become par for the course for these vendor’s (and others) offerings to the SMB market. Of course, they all mix and match features, some doing better than others, but in the end these are not simplistic RAID arrays. After all, who needs the most help: the well staffed, well trained large enterprise organizations or the shoe string, thinly staffed middle sized outfits struggling to meet their storage needs? What could be better then storage gear that takes the “rocket science” out of storage management? We are in a new era for business oriented storage products, with differentiation based on rich feature sets, pricing of those features, service and reliability separating the leaders from the laggards.

Gone are the days of a commodity, stripped down Yugo of a RAID storage subsystem. Look for and demand high function from storage subsystems – anybody can do the simple stuff – it’s the high valued, labor saving, simplification features that will really count in your organization.

Posted by Gene Ruth

Can a Cloud Get Ahead of Itself?

2008-08-05T07:42:25-07:00

The recent Amazon S3 outage has left me wondering if we aren't in danger of getting caught up in the incessant hype of cloud computing and deploying important applications on it before it's really ready. The end user impact of the S3 outage was bad enough, about 6 hours of downtime for S3, not good if you were doing weekly image backups using a backup service provider that leveraged S3 for storage! More troubling was the cause of the outage; corrupt control messages that rapidly turned into a firestorm of mutual recrimination between the nodes as they argued about the state of the cloud until finally the whole thing went to hell in a hand basket. All this happened because there was little or no error checking on the inter node message traffic so the corruption went unnoticed until it was too late to do anything but reboot the entire S3 infrastructure. With hindsight, the importance of protecting the integrity of the inter-node messages seems obvious, and I'm sure Amazon will fix that problem, in fact they may already have done so. What's really worrying is that a design flaw of that magnitude could ever have made into a production system hosting customer data, it begs the question, "what other surprises are buried in the code currently powering cloud computing applications?"

Don't get me wrong, I do believe that in time cloud computing in one form or another will be a key part of corporate IT, the question is when, not if. In the past, something as experimental as the "cloud" would have been hammered on in academia long before it became a part of IT infrastructure. But today, this type of development happens at breakneck pace with commercial companies (often relying on VC money which creates a requirement for rapid customer growth) claiming "enterprise readiness" for products that have barely advanced beyond the science project stage. In conclusion, I think the Amazon S3 outage should make IT organizations put the cloud on the back burner for the moment or at least limit it to the lab and non-essential applications. Let's give the vendors at least a couple of years to:

Shake the bugs out on non-mission critical services where 6 hours of unplanned downtime isn't a resume generating event (RGE).
Demonstrate their ability to deliver the continuous service needed for enterprise applications

Posted by: Nik Simpson

It's Official - The Hypervisor is now a Commodity

2008-07-22T15:51:54-07:00

On today's Q2 earnings call, new VMware CEO Paul Maritz announced that VMware will offer the ESXi hypervisor as a free download. This will clearly be perceived as a bold decision and one that will likely have some VMware investors reaching for their favorite antacid. In my previous post, VMware's Leadership Change - The Dust Settles, I commented that VMware should take immediate steps to significantly drop prices. To me, an ESX price drop is one of several moves that I think VMware has to make. VMware is increasingly seeing competition and in offering its core hypervisor at no charge, VMware has matched similar offerings from rivals Microsoft (Hyper-V) and Citrix (XenServer Express Edition).

Congratulations are definitely in order because I think VMware has made a good move. Taking their flagship hypervisor and offering it for free positions them well alongside their primary rivals and makes it easier for organizations with limited budgets to transition to VMware-based virtualization. To be clear, the free ESXi offering includes the hypervisor, VMware's virtual machine file system (VMFS), and symmetric multiprocessing (SMP) support. ESXi had been priced at $495 for a server with up to two processors. Most production virtualization deployments include high availability, and if you want to add high availability to the free ESXi hypervisor, you'll need to upgrade to a VMware Infrastructure (VI) Standard license, priced at $3,624 for two processors (with the required 1 year Gold support package). Upgrading to the Enterprise license ($6,958 for two processors [includes 1 year Gold support]) would get you features such as VMotion (live migration) and the distributed resource scheduler (DRS). The reason I'm bringing up upgrade prices is twofold. First, I don't think VMware is doing enough with regards to price drops. Second, since practically all VMware sales include the standard or enterprise VI tiers, the impact on revenues will be practically non-existent.

My colleagues who were at Novell at the time that Microsoft's Active Directory was attempting to take down Novell's eDirectory look back and note that Novell was far too conservative in its attempt to protect revenue streams and ultimately lost to Microsoft (somewhere in the VMware corridors someone is now shuddering at yet another Novell comparison). Giving away your flagship hypervisor is a bold move, but VMware really isn't taking much off the table. VMware's core revenues have long been driven by the features (e.g. HA, VMotion, DRS) it sells around the hypervisor. In my opinion, I think VMware needs to do more than give away its hypervisor if it hopes to hold back competition. Instead, VMware should aggressively take sales opportunities from its competitors. To do that, I think VMware needs to drop the prices of its VI Standard and VI Enterprise packages by at least 33%. Chopping a third off the price of its core packages would make its pricing comparable to its main rivals. For example, the Stratus OEM of Citrix XenServer with high availability is just under $2495 per physical host. A comparable solution from Virtual Iron is priced at $799 per socket. If price is equal, purchasing the hypervisor with the most features becomes a "no brainer."

VMware will say that there's more value in their hypervisor, and it's an enterprise-proven platform. So it's right to say that a higher price is warranted. However, a price that is often 2x-3x higher than competitive solutions is putting IT decision makers in a very tough spot, especially as alternative solutions that provide essential virtualization features such as high availability are emerging. Also, VMware must carefully weigh the potential backlash from existing customers who purchased software at a higher price or who are locked-in to a lengthy enterprise license agreement (ELA). Still, to me the question is not "What's the pricing model that keeps us competitive today and allows us to maximize revenues?" Instead, the question should be "How do we leverage pricing to deny market opportunities for our chief competitors?" If the question is the latter, VMware needs a significant price drop in both their VI Standard and VI Enterprise packages. A large drop would open more doors for VMware and remove price as a factor in many virtualization purchasing decisions. What VMware gives up in short term revenue would be recuperated in long term licensing and maintenance contracts. Virtualization is an infrastructure technology, and as such, has higher barriers to exit - organizations with substantial virtualization investments will not rip and replace it on a whim. So today's game is all about penetrating the significant untapped virtualization market. Yes - VMware's already doing that today, but I believe they could be more aggressive. Lower prices today results in increased market penetration and increased long term software maintenance revenue.

In addition to staying price competitive, VMware needs to hit back against Microsoft. Making ESXi free has effectively placed the hypervisor as a commodity, and as a result VMware has taken Microsoft's bait. VMware now needs to hit back. I alluded to some of the ways VMware can do this in my previous post. We've had quite a bit of internal discussion on this topic at Burton Group as well, and I'd love to hear your thoughts on what VMware's next move should be in their chess match with Microsoft.

A conservative, complacent pricing and licensing strategy will not be enough. Sure, that's the "textbook" approach to running a software company. Microsoft loves the textbook. After all, their competitors historically use textbook management practices in their attempts to take down or hold off Microsoft. How well has that worked? VMware - throw away the textbook. Paul Maritz has the pedigree to write a best seller on software vendor strategy. We remember best sellers because their text often defies conventional wisdom and surprises us along the way. A conservative price drop is a good story, but not a best seller.

Posted by: Chris Wolf

VMware's Leadership Change - The Dust Settles

2008-07-17T11:25:29-07:00

Last week while my colleagues were debating the impact of Paul Maritz replacing Diane Greene as VMware's CEO, I was relaxing on a beach in North Carolina's outer banks. I prefer to be "unplugged" while on vacation, so I didn't hear of the news until I returned home last Saturday. My colleagues Drue Reeves (see VMware: Welcome to the Game) and Richard Jones (see VMware CEO Ousted) did an excellent job covering the leadership change, and after having spoken to Paul on Tuesday, I think it's a good time to weigh-in with my thoughts regarding VMware's executive leadership change.

Diane Greene, in my opinion, did a tremendous job in bringing VMware to where it is today. VMware ships the industry's gold standard for x86 server virtualization, has strong sales, and is continuing to push the innovation envelope. This all happened on Greene's watch. In addition, Greene had VMware strategically positioned to continue broadening the company's growth and market presence. With that being said, it shouldn't be surprising that folks inside and outside of VMware felt that a leadership change at this point was a bad move. Alessandro Perilli uncovered much of this turmoil with the following posts on Virtualization.info:

Complacency has never been an effective strategy against Microsoft, so I must say that I understand the timing of the move. VMware has to move forward with a sense of urgency, and dare I say desperation. A complacent VMware would likely get steamrolled by Microsoft. History has proven that time and time again. NetWare and Word Perfect are two good examples of best-of-breed products that eventually succumbed to the Microsoft juggernaut. As Microsoft's previous successes have shown, they don't have to have the best technology to win. "Good enough," combined with low total cost of ownership and simple, effective management has historically worked very well.

Now let's get back to the leadership change. I see it as a good move. Many don't like Maritz's lengthy history with Microsoft, but that's exactly why I see him as the right leader at the right time. Maritz was part of Microsoft's core executive team that took Microsoft from dismissed start-up to dominant force in office applications, desktop operating systems, and server operating systems. Maritz knows how to take on and defeat large companies with well established product lines. Yes - VMware has the dominant hypervisor today, but VMware's future depends on much more than hypervisor dominance. VMware's future will ultimately be tied to the value it brings to applications.

At the end of the day, what matters most is whether or not our applications are available to users. Microsoft realized this long ago, and when you look past virtualization, Microsoft has a better application management story than VMware today. VMware has the best virtualization management solution going, but in the end I don't see that as being enough to hold out against Microsoft. I mentioned in my Apple - Opportunity is Knocking post that the desktop as we know it is in a state of transition and opportunity exists for a more innovative solution to eventually supplant Windows desktop operating systems. Five years out, the desktop operating system will be fundamentally different. The same can be said for the server, as the OS is trending toward being a thinner, modularized foundation for running applications. Just look at Windows Server 2008 core and the success of Linux-based virtual machine appliances to see that. In addition, we'll be seeing some core IT services transition to cloud-based computing. The cloud may be external to the organization, and I believe many organizations will have their own centrally managed clouds as well (essential to meeting the organization's security and compliance requirements).

Nearly all elements of the traditional data center are in flux, and while this is exciting, it has to worry Microsoft. Microsoft sees the transition and is developing their own next generation application, client, and server solution set. For VMware to win, they will have to deliver a fundamentally superior solution to Microsoft, and do it sooner. Of course, this will initially be a complement to existing Microsoft products. But if VMware has its way, many core IT services will transition to cloud-based offerings and virtual appliances. In the mean time, it's clear that Microsoft isn't sitting on it's hands. If you want proof, take a look at the Live Mesh Technical Preview. Live Mesh is a very intriguing cloud-based desktop delivery solution and should have VMware's attention.

As an analyst, I can afford to be wrong once in awhile and still keep my job. Maritz, on the other hand, likely doesn't have this luxury. That being said, if I was in Maritz's shoes, I would immediately do the following:

Lower the prices of the entire VMware product line
Accelerate development on a soup-to-nuts solution for the SMB space
Accelerate development on VMware's virtual desktop solution
Focus the company's messaging around the application and the total solution

Microsoft and Citrix are both competing against VMware on price. If VMware refuses to lower prices, vendors such as Microsoft, Citrix, Virtual Iron, Novell, and Sun will gladly take the market share that VMware concedes. If VMware wants to take away those market opportunities, then it must drastically lower prices immediately. Short term, a substantial price reduction would hurt. However, over time, denied opportunities for VMware's competitors would equate to more sustained revenue for VMware.

Microsoft has long dominated the SMB market and feels strongly that Hyper-V will do well there. When you factor in the System Center management suite and Data Protection Manager, by the end of the year Microsoft will have a very good management and backup solution to complement Hyper-V (System Center Virtual Machine Manager 2008 has yet to ship). Microsoft is saying that Hyper-V runs on "the Windows you know and love," but if you configure Hyper-V in the recommended deployment on Windows Server 2008 core, I'm not sure how much love you'll be feeling immediately. I've found a Hyper-V on Windows Server 2008 core installation to be more complex than either VMware ESX Server or Citrix XenServer. VMware should capitalize on this, and Microsoft should counter by simplifying the Hyper-V/Windows 2008 Core installation process (yes - it's easy to enable the Hyper-V role, but setting up the network, storage, domain membership, etc. on Windows Server 2008 Core will take additional time). To win the SMB, VMware needs to be cost competitive and offer equivalent SMB management features. I think adding a solution that rivals Data Protection Manager would help.

Some see the virtual desktop as an irrelevant tangent, but not me. Microsoft eventually won the server war due to its integration with the desktop, and I see virtualization as playing out similarly. Citrix is gaining a lot of momentum in the virtual desktop space and Microsoft is telling anyone who will listen to go with the Citrix solution. So VMware has an uphill battle here. VMware has some great ideas for virtual desktop delivery and I feel that they will need them on the street sooner rather than later in order to keep Citrix at bay. Losing the virtual desktop battle may not cost VMware the virtualization war, but it will definitely make their road to success all that more difficult.

Microsoft's core messaging has focused on the solution, while I feel that sometimes VMware places too much emphasis on its individual technologies. When you have a best-of-breed product, you tend to want to brag about the little things that you can do that no one else can. So I understand why VMware is proud of it's individual products. Still, what I want to see from VMware is more about how they are adding value to the entire application stack (management, availability, performance, etc.). Sure, VMware doesn't do it all, but combined with its major partners in the management space, VMware plays a major role in the complete solution. Give us a solution web page that shows how you do it. Better yet, give us a web 2.0-based site that allows anyone to design a fully virtualized data center using products from VMware's core management partners (e.g., BMC, CA, Dell, IBM, HP). The best technology doesn't always win out in the end, but the best story often does.

VMware - tell us your story.

Posted by: Chris Wolf

Storage virtualization – whatever you want it to be.

2008-07-14T14:15:22-07:00

Storage virtualization - what is it? Does anyone know? I hear the term thrown around describing all kinds of storage devices. It must be good - after all who doesn't love to talk about 'server virtualization'. Kinda the same thing with 'green storage' whatever that is - but that's another blog. Virtualizing anything must be good nowadays - what vendor doesn't want to be associated with the server virtualization stampede, and rightly so.

The definition of storage virtualization according to the Storage Networking Industry Association (SNIA) includes terms like: abstracting, hiding, aggregating, isolating devices, or the delivery of new services. With that definition, pretty much anything can be described as being virtualized. Cool. So if I'm a vendor and I offer a RAID device, then by definition my device supports virtualization- call marketing and get that on the flyer for heaven's sake.

The whole storage virtualization lingo has gone astray - it's become meaningless with overuse and nothing more than a marketing jingle. Now I'm not saying that the functions offered by virtualization are not useful, they definitely are, but let’s call it what it is or should be: 'storage simplification'. An ultimate user should be able to identify what app's she plans to run and order up a chunk of storage to meet that need. And if her data needs protection then order that up too. Does the user or IT admin really want to muck around with the parts and pieces in a storage system to meet the needs? No. It’s much kinder to simplify the complexity and automate by simplification whatever is needed under the covers. It’s the 'whatever is needed' part that storage virtualization vendors focus on. I push the gas pedal the car goes - I don't adjust the timing and fuel mixture - the gas pedal simplifies that for me.

So what's important? Forget about the term 'storage virtualization'. Focus on function and features. Determine what you are looking for in the ultimate storage ecosystem. Do you want to provide nuts to bolts information lifecycle management? Ok then, look for storage vendors who can move data auto-magically from storage tier to tier with some content management thrown in. Or are you tired of managing volume capacities? Then get yourself into thin provisioning. Need volumes to follow around server virtual machines? Look for storage that can flit along with the VM. Need to backup and archive? Get data deduplication. Consolidate all those disk arrays you've got into a common management point with a few extra features thrown in? Yes we can. Whatever - you get the idea. Call it what it is, describe the function you're looking for and find vendors who can deliver. If the storage vendor wants to call it virtualization, then great, give him a knowing smile, pat him the back, but ask for a firm bid.

Posted by: Gene Ruth

VMware: Welcome to the Game.

2008-07-09T13:03:51-07:00

Now that the initial shock of Diane Greene's resignation has begun to subside, the "what happened" question begins to loom as people try to understand what is going on at VMware -- a company that has had the biggest tech IPO since Google.

The elephant in the room is whether the relationship between Diane Greene and Joe Tucci (EMC CEO and VMware chairman) was strained. I want to put that aside for a second and talk about the real issue that perhaps contributed or even led to yesterday's events... because whatever your feelings are about the tension between the two, it is secondary (at least) to the real problem -- competitive strategy in an increasingly competitive market.

Game On
The primary issue is VMware's competitive strategy against Microsoft Hyper-V, Citrix XenServer, and other XEN hypervisor players. VMware’s fundamental competitive problem is very analogous to Novell’s struggle against Microsoft in the 1990's. Novell had NetWare and NDS (Novell Directory Services) two of the strongest "system software" products in the market. But Novell didn’t have the enterprise applications that could keep customers from switching to a non-Novell centric environment. Microsoft did. Eventually Microsoft leveraged their application position to take market share away from NetWare by tying applications like Exchange to Windows (and Active Directory). VMware faces almost the very same issue. VMware has ESX -- a superior hypervisor in many respects to Hyper-V. But one has to ask: how long will it be before Microsoft ties Hyper-V to Exchange, SQLServer, or some other part of the Microsoft ecosystem? Hyper-V may not be the best hypervisor in the market, but it might be the best hypervisor for Exchange or for Windows environments. And given Hyper-V’s price (free), it might just be “good enough” to compete (remember -- Active Directory wasn’t near as feature rich as NDS, but because Microsoft tied AD to Exchange, AD is now the defacto directory service ).

The “hyper-V effect” on ESX – at least at this stage – was more of a “pause” than market share shift. Once again, Microsoft pulled off their typical (yet somewhat genius) MO by announcing vaporware, thereby buying time in the market, and then releasing a product with less features than the market leader, at a significantly cheaper price. Now the game ensues again….ESX will see a “bounce” in the sales because consumers will realized that Hyper-V isn’t quite there (yet). But Microsoft accomplished their first goal...get IT administrators “kick the Hyper-V tires” in their labs and gain entry into the market. Be assured that Microsoft will continue to add functionality to Hyper-V and eventually tie their hypervisor to their enterprise applications (like Exchange and SQLServer) and incorporate Hyper-V management into System Center. Eventually one has to wonder if Hyper-V will be “good enough” for customers to migrate because hyper-V works well within the Microsoft ecosystem. I suspect VMware's lack of competitive strategy against such forces was a central cause for concern in the boardroom.

Welcome to the game VMware.

Winning Strategies
Now, I'm not saying VMware is going away. Far from it. They are the market leader and will continue to be in a rapidly growing market. But, they have to address the competitive strategy issue if they want to continue to lead, because right now, they're playing Microsoft's game...and Microsoft is very good at it.

So what should VMware do? How do they address this issue I've raised? A winning strategy is multi-pronged.

VMware should stop focusing (so much) on the functionality of the hypervisor and start focusing on strategies that make the OS "irrelevant". If the OS running inside a VM is simply an "operating runtime environment" for the application, then it doesn't matter "what" that runtime environment is...in other words, it's irrelevant (although necessary). This strategy marginalizes all operating systems...including Windows and Microsoft's application position. VMware has a clear lead in hypervisor technology (perhaps 2 years or more). While they can't let their hypervisor slip, there is only so much functionality one can add to that layer. By strengthening their application partnerships with ISVs like SAP AND demonstrating that these applications don't need a “bloated OS” to run, VMware might be able to convince customers that running Windows in a VM is overkill. Do applications really need all of the functionality contained within Windows to run a single application? No. So the rest of the "services" loaded by Windows are simply chewing up memory and CPU utilization. VMware needs to exploit this fact, by pressing their "JUICE" initiative or Just enough OS (JeOS). The idea is to load "just enough" of an operating system to run the application.
Some have suggested that VMware should focus on management, and that, could be a place where they would "win" against Microsoft. While I can agree that management is a differentiator and a place where VMware can compete there are limitations. VMware providing a better management tool for virtualization is -- in many ways -- simply another management silo within the context of the overall IT ecosystem. Microsoft, with System Center, is better positioned to manage the entire IT ecosystem -- everything from operating systems to applications to storage and servers to virtualization. Do we expect VMware to manage email, storage, physical servers, or networks? Perhaps, but that's a stretch. It’s more likely that Microsoft (or some well-known enterprise management console) will win that battle. In addition, management is not something easily commoditized, so customers have the expectation to pay for what they get. Whereas the hypervisor is being economically marginalized. This was clearly Microsoft’s plan from the beginning, given Hyper-V's price. So, VMware can't "go it alone" in the management space and expect to win in the long run. In the management arena, a better strategy would be for EMC and VMware to combine forces. It's no secret that EMC wants to be a major player in the management space. Teaming up, VMware and EMC could create an enterprise management suite that could manage virtual servers and desktops, virtual storage, and SANs. Partner that with Cisco, and you have the makings of a very nice enterprise management offering.
VMware should "own" the cloud. Every hosting and service provider in the market should be selling applications -- or SaaS -- using virtual servers running ESX. Make ESX a common part of how companies use the Internet services and resources from the cloud...that's very sticky and hard for people to replace. In addition, if VMware can provide the best management experience for the service providers AND for the customers using the resources from the cloud, then VMware can own the market. It also allows VMware to quitely replace applications like Exchange with other email applications that provide the same email "service" to customers. Does email as a service require Exchange? or Windows? I suspect that new CEO Paul Maritz will focus in this area.
VMware should accept/acknowledge that Hyper-V will exist and extend ESX into the enterprise. WHAT?!!? In other words, recognize that -- in the long run -- people are going to have Hyper-V, so make Hyper-V just another part of a heterogeneous enterprise where multiple hypervisors exist. ESX is the "enterprise hypervisor" and Hyper-V is less capable one, but good enough for some applications. However, no one manages -- or virtualizes -- a heterogeneous environment better than VMware. Heterogeneity is something Novell did very well, and continues to do today. Sounds anti-intuitive? Look, the road to perdition is littered with the bodies of companies (WordPerfect and Netscape come to mind) who tried to stand in the way of Microsoft products and lost. VMware should not do the same. If VMware strategy is to compete in the hypervisor space, then they're playing the wrong game.
Diversify in the virtualization space. VMware is doing this to some degree, but one can make the argument that their strategy has been to focus on server virtualization (that's why they have the best product). VMware needs to increase the scope of their offerings and the focus and attention they pay in those areas.

Maybe VMware is doing these things already. If so, then time to press on the accelerator. If not, time to get going...and fast. For VMware, there's no better time in the market to exact this strategy. Vista sales are lackluster and the transition to Windows Server 2008 may prove slow. In addition, Microsoft is busy fighting another battle that threatens their OS and application dominance: Google apps.

New CEO
I've seen some comments on blogs that suggest VMware -- with the hiring of Paul Maritz -- will cozy up to Microsoft....that VMware should forge the same relationship that Novell has with Microsoft to ensure platform interoperability. Maybe. Interoperability is a good thing for customers. But I’m skeptical this is a comprehensive winning strategy. Whenever a company does this, it signals that the battle between the two is really over. The “if we can’t beat them, then join them” strategy is one where Microsoft ultimately wins. What VMware really needs is a competitive strategy. Given Paul's recent startup in the cloud computing space it seems likely that VMware may be headed in toward the cloud computing space. Perhaps this is a smart move, as it adds to the "OS is irrelevant" strategy. ...it is certainly the right game to play.

posted by: Drue Reeves

VMware CEO Ousted

2008-07-08T09:25:13-07:00

VMware chief executive officer Diane Green is ousted and x-Microsoft leader Paul Maritz is put in place.
VMware stock price sank on the news (who would have guessed?)
Back in January, I blogged about the troubled times I saw ahead for VMware; making a comparison to Novell's rise and fall. Similarities continue to abound. Remember Bob Frankenberg being ousted from Novell? Remember Novell brought in Eric Schmidt (now Google's Chief) to save the day?
VMware has adjusted its earning forecasts lower for the remainder of the year. Wall Street never likes it when a company does that, and they are feeling the pain. There is a relationship between the company's slowing growth and Ms. Green's departure. The board needs the company to change its focus and direction as the virtualization market matures. As BurtonGroup has stated, 2008 is the year of choice in virtualization solutions in the x86 commodity server world. Xen solutions from virtually (pun intended) everyone and their dog, Hyper-V from Microsoft, not to mention the slowing economy is spelling a drastic need for change at VMware. VMware's board is looking for that change of focus in appointing a new CEO.
My experience has been that the success of a new CEO coming on board is 30% the CEO's abilities and 70% the rest of the employees in the company being able to take a step back and recognize the changing market around them such as changes in what's important to users and focus their development and go-to-market efforts around those changes.
As I mentioned in my January blog, VMware has great opportunities; they just need to act on them and quickly.
[Posted by Richard Jones]

SSD - Storage is just the first wave

2008-07-08T08:18:12-07:00

Solid state disk (SSD) made from FLASH memory technologies are suddenly all the rage, with big vendors like EMC and SUN giving details on their strategy for using SSD to improve I/O performance in enterprise storage. It's not surprising that storage is the first to benefit from the technology for two reasons:

Performance: Conventional disk cannot address the growing gap between storage and overall system performance. You can't just make disks spin faster or the read/write heads move more quickly, physics gets in the way.
Ease of use: FLASH-based memory packaged as disk (i.e. using the same 3.5" and 2.5" form factors and interfaces) is pretty easy to integrate with existing storage systems, it doesn't take a rocket science degree.

The immediate impact will be on sales of high performance SAS and FC disk, where performance is prized over capacity. Why buy (and subsequently power and cool) ten 72GB 15K RPM FC drives if one or two SSD drives can give you better performance and still meet your capacity requirements? Yes, I know the technology isn't quite there yet, particularly for write performance, but it will be, and it will decimate the high-performance disk market.

With SSD in disk drive form factors a foregone conclusion (my bet is most enterprises will have it for at least some applications by 2010), the more interesting question is how the technology evolves and it's long term impact on several areas:

Server form factors: Some of the form factor of a typical server or blade is governed by the number of drives supported. But why have drives at all, FLASH memory doesn't have to be packaged that way, it could installed in slots on a motherboard for example. So FLASH bring about new more dense server form factors as storage performance, power consumption, and storage form factors change.
File systems: All the file systems in widespread use today are designed to overcome or at least mitigate the inherent limitations of conventional disk. For example many file system optimize placement of disk blocks for a particular file to minimize disk head movements caused by fragmentation, others employ defragmentation tools to address the problem. But FLASH-based storage is truly random access, performance is the same for every single block on the device, rendering the whole concept of fragmentation meaningless.
I/O interface protocols: Protocols like FC and SAS are designed for use with conventional disk, but they don't make sense for high-performance I/O from memory, which would be better suited to a pure memory type interface without all the storage protocol overhead.
Main memory: At first glance this one doesn't seem that attractive, after all FLASH is much slower than DDR2 DRAM isn't it? But it doesn't have to be that much slower, as Spansion are demonstrating with their NOR FLASH-based DIMMs that plug right into the memory slots on the motherboard. Read performance is not quite up to DDR2 speeds, but it's close enough, and 32GB/DIMM (that's 256GB in eight slots) is very attractive for applications that primarily read from a database that can now be loaded into memory.
Operating systems: Changing the performance ratio between storage and compute may open up new possibilities in OS design, for example the paging system could be implemented on FLASH storage interfaced directly to the memory bus.

I could go on (and probably will do in the future :-), but that will do for now.

Posted by: Nik Simpson