Energy lab gets three midsized Linux clusters for nonclassified simulations
Featured eBooks
Connecting state and local government leaders
An Energy Department laboratory is collaborating with two vendors on fast-track construction of three high-performance clusters running Linux.
An Energy Department laboratory is collaborating with two vendors on fast-track construction of three high-performance clusters running Linux.SGI's federal division and Linux Networx Inc. of Sandy, Utah, are building the clusters from 472 1.7-GHz Intel Pentium 4 Xeon processors. They will process Lawrence Livermore National Laboratory's midsize simulations that don't require classified systems with thousands of nodes, said Mark Seager, the lab's assistant department head for tera-scale systems.Together, the clusters will cost $1.7 million, which Seager called an 'outstanding price' that includes three years of hardware and software maintenance.The three clusters will contain 126, 86 and 24 dual-processor nodes, respectively. All three will have 2G of Rambus dynamic RAM per node and two Linux Networx systems management applications, ClusterWorx and Ice Box.The OS will be Red Hat Linux 7.1 from Red Hat Inc. of Durham, N.C.The Accelerated Strategic Computing Initiative's IBM SP supercomputers, White and Blue Pacific, handle the most complex simulations of nuclear weapons aging for the National Nuclear Security Administration. It's difficult to load the massive systems with large as well as small jobs, Seager said, so Livermore officials wanted smaller clusters for more efficient scheduling.The deal was signed Aug. 24, and lab officials expected to complete the clusters last month. They will transfer the two larger clusters to the lab's classified environment, Seager said.Lab officials made the time frame tight because they wanted to beat year-end budget closeouts, Seager said.'We think it's quite doable,' he said.The 48-processor cluster will remain unclassified so that the lab and the two vendors can use it for open-source development projects.Although scientists at Livermore, Los Alamos National Laboratory and Sandia National Laboratories all use the ASCI White system, the smaller clusters will be more for local Livermore users.Internal switches from Quadrics Supercomputing World Ltd. of Bristol, England, will transfer more than 300 megabytes/sec between nodes, Seager said.The 252-processor cluster will have a theoretical peak speed of 857 billion floating-point operations per second, said Wayne Vieira, SGI's on-site systems engineer.Linux Networx is providing the hardware and SGI Federal the engineering work, said Kathi Cloutier, SGI's account manager for the lab. SGI Federal will keep an on-site engineer at Livermore for a year to assist with the cluster.All three Energy labs that participate in ASCI have ongoing Linux cluster projects, Vieira said, citing Sandia's Cplant [] and Los Alamos' Loki project.
Smaller clusters will let Livermore facility more efficiently schedule small jobs
Wearing Red Hat
Local traffic
GCN, July 23, Page 5
Lab officials made the time frame tight for delivery of the cluster because they wanted to beat year-end budget closeouts, Mark Seager said.
Wearing Red Hat
Local traffic
GCN, July 23, Page 5
NEXT STORY: Storage device stacks rackless