Cray Tweaks CS300 For Large Memory Workloads

Cray creates two new configurations of its CS300 machine to cope with high-memory workloads

Supercomputer maker Cray has partnered up with ScaleMP in order to create two new configurations of its CS300 compute cluster

The move adds large memory and shared memory solutions to its portfolio.

New Configurations

Cray will leverage ScaleMP’s vSMP Foundation high-end virtualisation technology in the new solutions – the CS300 LMS system and CS300 SMP – that will enable end users to take jobs that require high levels of memory and run them on virtual machines that can handle such workloads as complex modelling analysis and big data tasks.

Cray Titan topScaleMP’s software lets the applications view a number of x86 systems as a single virtual system, enabling organisations to leverage the virtual machines for the jobs rather than having to invest in traditional SMP hardware systems, according to Cray officials. The systems also eliminate the need for users to buy separate cluster management software or to update multiple operating systems and applications.

The preconfigured CS300 systems “contain massive amounts of memory and large numbers of computer cores, and are also easy to deploy, manage and service,” Daniel Kim, senior vice president and general manager of cluster solutions at Cray, said in a statement. “Researchers and engineers in a wide array of scientific domains with needs for large scale analytics and complex simulations can now take advantage of these very powerful systems.”

Cray Partners

Cray inherited the CS300 series of supercomputers in November 2012 when it bought rival Appro for $25 million (£15.7m). The systems come with an integrated high-performance computing (HPC) software stack and software tools that are compatible with most open-source and commercial compilers. Company executives have since been bolstering the capabilities of the offerings, such as when they announced in June that they were going to launch cluster supercomputers for Intel’s Hadoop distribution, creating a turnkey infrastructure using Intel’s Hadoop, a Linux operating system and Cray’s Advanced Cluster Engine (ACE) management software.

The results were systems that can better leverage Intel’s Hadoop distribution for big data jobs in HPC. Now the company is looking to longtime partner ScaleMP to bolster the CS300’s capabilities in running HPC workloads that require a lot of memory and can be managed as a single system.

The CS300 LMS – a large memory system – offers fast access to high-speed memory for such jobs as data analytics visualization for scientific research, engineering design simulations, weather simulations and bioinformatics, according to Cray officials. The system, designed for high-RAM requirements, offers direct access to many terabytes of memory without needing large numbers of processors. It scales from 4.75 to 8.75 terabytes of memory, and uses 20 to 32 Intel Xeon cores. Systems start at under $200,000 (£125,000).

The CS300 SMP, which is a shared memory parallel system, can run hundreds to thousands of workloads at the same time, due to the large amounts of processor cores and memory. The system includes 360 Xeon cores and 4.75TB of memory, with an optional upgrade to 8.75TB and 680 cores. The base price of the CS300 SMP is less than $300,000 (£188,679).

Both systems are based on air-cooled configurations of the CS300, and include hot-swappable, redundant high-efficiency power supplies. They are installed in standard racks that offer optional 480V power distribution and 208V or 277V three-phase power supplies. There also are options for chilled rack cooling, with rear-door heat exchangers providing more cooling capabilities.

How well do you know the cloud? Take our quiz.

Originally published on eWeek.