Sign up for our newsletter and get the latest HPC news and analysis.

Cray and virtualization provider ScaleMP hook up

From Cray’s website this morning, news that Cray is partnering with ScaleMP to use virtualization technology to bring shared memory to the company’s CX1 deskside super

Cray logoGlobal supercomputer leader Cray Inc. (NASDAQ: CRAY) and ScaleMP, a leading provider of virtualization solutions for high-end computing, today announced a strategic alliance to offer joint solutions based on the Cray CX1(TM) deskside supercomputer and ScaleMP’s vSMP Foundation. Available immediately, the joint solution will target the High Performance Computing (HPC) segment allowing customers to operate a shared-memory, deskside supercomputer that scales up to 128 cores and 1TB of shared memory.

…vSMP Foundation aggregates multiple industry-standard, off-the-shelf x86 servers (rack mounted or blade systems) into one single virtual high-end system for the HPC market. vSMP Foundation provides customers with an alternative to traditional expensive symmetrical multiprocessor (SMP) systems and also offers simplified clustering infrastructure with a single operating system. It currently allows customers to create a single virtual SMP system with up to 32 sockets (128 cores) and up to 4 TB of shared memory in an energy-efficient, dense package.

I will be very interested in the performance of this (and the other solutions) that are coming to market right now around software-enabled shared memory. If these solutions start to be performance competitive with hardware-supported solutions like SGI’s (or, more likely, they are able to use the virtualization hooks that the chip vendors are building into silicon to accelerate the solution) then this could end SGI’s shared memory business.

Comments

  1. Homer Simpson says:

    Many people who like SGI CC-NUMA machines run MPI codes and prefer the SGI’s for their reliability and excellent interconnect latency. HPC Centers turn on cpusets anyhow and limit the full use of the SMP capabilities. ScaleMP and other software only solutions like RNA Networks may be good for use as a ramdisk since that would likely still be faster than disk I/O. Hardware/software combo’s like 3Leaf will be better in cache coherency but will not address the underlying latency advantage that SGI has.

  2. Menelaus says:

    Please tell us the “other solutions” Thanks.

    I will be very interested in the performance of this (and the other solutions) that are coming to market right now around software-enabled shared memory.

  3. If you want HPC, go to Infiscale.com and get Perceus. Or if your not locked into your OS, download their Caos NSA Linux that comes with Perceus, Warewulf, ofed, openmpi, etc and you have a cluster up in a few minutes. We were running HPL runs in 15 minutes on 512 nodes. The power of open source. :)

Resource Links: