Cluster Monkey

Intel Splits The Network

deadline@clustermonkey.net (Douglas Eadline) — Fri, 19 Apr 2013 11:53:02 -0400

In a good way, we think

Intel has recently announced a new Ethernet Open Network Platform that splits ("disaggregates") the control plane from the data plane and provides users the ability to control network aspects that were previously hidden "inside the box." The idea is to create global control of large networks, rather than relying on local switches. The control processors are of course of Intel x86 variety. There is an SDK called the Data Plane Development Kit (includes kernel by-pass tools) so that users can twiddle with the network design at a low level.

Don't expect to see products right away, but these are open reference designs offered by Intel. They may offer some interesting options that allow networks designed and tuned for HPC. There is a good in-depth write-up at SemiAccurate.

Experiments With Switchless 10GigE: The Bonded Loop

deadline@clustermonkey.net (Douglas Eadline) — Thu, 11 Apr 2013 12:27:35 -0400

When the best solution just won't fit the box and the budget

Modern Ethernet technology is based on network adapters and switches. Using Ethernet without a switch only happens in rare situations where a small number (e.g. two) systems need to be directly connected together. Such a connection is often called a "cross over" connection because a special cable may be needed.

The cost of adapters and switches follow a very predictable commodity pricing trend. At first the cost of systems is quite high and decreases as the sales volumes increase. Currently Gigabit Ethernet (GigE) enjoys low cost and wide availability from multiple vendors. Ten Gigabit Ethernet (10GigE) is now experiencing greater acceptance and thus decreased costs. Although volumes are growing, 10GigE still commands a high per port price (Adapter/Switch) and thus can be an expensive option for many small projects.

Scalable Informatics Storage Systems at SC12: Fast, Faster, Fastest

deadline@clustermonkey.net (Douglas Eadline) — Tue, 19 Feb 2013 16:19:05 -0500

There is fast, then there is "Scalable Fast"

Scalable Informatics (SI) makes fast storage systems. How do we know they are fast? Because SI provides the numbers upfront directly from the command line in founder Joe Landman's blog; Scalability.org. This year at SC12, SI was showing remarkable numbers, both in capacity and performance, for some of their products.

To give you an idea, consider the their JackRabbit JR4 60 unit. It is a 4U storage device that can hold up to 60 HDD's which translates in to 240 TB or data (that is just about a quarter petabyte of storage). In addition to high density, the system is specified at 7,000 MB/s read and write throughput. That works out to reading a 1 TB file in 144 seconds. You can see some real benchmark numbers on Scalability.org.

Intersect360 Research Hires Michael Feldman as Senior Analyst

deadline@clustermonkey.net (Douglas Eadline) — Mon, 11 Feb 2013 16:31:22 -0500

From the notable news department

Some recent news that arrived in my in box. Intersect360 Research, a leading market intelligence, research, and consulting advisory practice for the High Performance Computing industry, announced today that it has hired Michael Feldman, a 35-year computer industry veteran, to augment its analyst team.

Feldman, recognized as an expert in the HPC industry worldwide, is well-known as the former managing editor of HPCwire, where he spent eight years as one of the foremost predictors of HPC trends. Feldman was also the co-host, along with Intersect360 Research CEO Addison Snell, of the insightful and popular weekly HPCwire Soundbite podcast.

Numascale Delivers Plug and Play SMP at SC12

deadline@clustermonkey.net (Douglas Eadline) — Wed, 16 Jan 2013 12:28:20 -0500

Delivering big SMP performance at a cluster price

This year at SC12 Numascale, maker of low cost plug-and-play high performance shared memory systems, was showing performance numbers and some happy clients. For those that may not know, Numascale produces the NumaConnect adapter that provides plug-and-play connection of standard servers running a single un-patched OS version. Applications have fast access to the aggregate memory of all servers in the system using a hardware controlled ccNuma environment.

NumaConnect facilitates very large shared memory systems built from commodity servers at the price level of high-end clusters. This combination invites a new level of price/capability/performance that has never before been demonstrated using pure hardware. Systems built with NumaConnect run standard operating systems and all x86 applications.