RealTime IT News

Pentagon Clusters Around Linux

Ron Miller

The Pentagon will use a 2132 processor Evolocity II cluster computing system from Linux Networx to equip the Army Research Laboratory Major Shared Resource Center (MSRC), the Department of Defense (DoD) announced Friday.

The cluster, which was purchased under the High Performance Computing Modernization Program, is the largest ever sold to the DoD to make use of new Intel 64-bit extension technology. Financial details of the deal were not disclosed.

Due to be completed later this year, the cluster will have 1,066 nodes with dual Intel Xeon 3.6 GHz processors (2132 total processors).

The MSRC conducts intensive research and needs an efficient high performance computer to complete their research efforts.

"The work is physics-based computation -- flow dynamics, computational structural dynamics, computational chemistry -- that takes hours and hours and sometimes weeks to complete," Charles J. Nietubicz, acting deputy director, Computational & Information Sciences Directorate for the ARL, told internetnews.com. "These are complex equations involving millions of computations that you need a super computer to solve those kinds of problems,"

Clusterworx 3.0 and ICE Box management tools from Linux Networx will be deployed on the cluster to provide cluster management from one interface. The cluster will also use the Myrinet high-speed interconnect and Gigabit Ethernet technology from Foundry Networks.

"The DoD's adoption of cluster technology for the Technology Insertion program continues to validate our technology's reliability and high performance computing capabilities," said Dean Hutchings, COO of Linux Networx, said in a statement.

One reason Linux Networx has been able to achieve this level of computing power is the evolution of Intel processors combined with their ability to put together a reliable package of software and hardware.

"Intel has gotten very fast, and Linux Networx has wrapped around them a very efficient system. In the past, if you had a 2000 processor system, you were lucky if you had 100 hours of use before anything went wrong," said Earl Joseph, vice president of high performance computing at IDC.

Because a "lot of reliability" has been built into Intel hardware and software, Joseph added, "Their machines won't crash every 100 hoursIf one processor goes down you only lose the processors on that node, but whole machine doesn't go down, so only one part of the application needs to be restarted."

Eric Pitcher, vice president of product marketing for Linux Networx, said the key to cluster productivity is software.

"We are providing end product to the customer that includes cluster management software with a whole host of productivity features such as resource management and scheduling along with a whole host of features to help monitor the system," he said.

Last year, Linux Networx delivered a 256-processor cluster to ARL.

"In a sense, the last year has been a proving ground, and based upon that, they made decision to go for larger system. They didnt jump into deep end of the pool. They started in shallow end and worked their way out," Pitcher said.

The Linux Networx cluster was purchased as part of the Technology Insertion 2004 (TI-04) program, an initiative to modernize the Department of Defense high performance computing (HPC) capabilities. The HPC provides the supercomputer services, high-speed network communications, and computational science expertise that allows the U.S. Defense laboratories, such as ARL, to conduct a wide range of focused research, development, and test activities.

As part of the TI-04 program, Linux Networx will deliver five other cluster systems to other DoD Centers. Details about the five other DoD HPC cluster systems will be released later this year.