Architectural requirements for the Red Storm computing system [report]

William J. Camp, James Lee Tomkins
2003 unpublished
Albuquerque, NM 37 185-1 109 P.O. BOX 5300, MS-1109 Abrtract This report is based on the Statement of Work (SOW) describing the various r e q u i m n b fm delivering 3 new supercomputer system to Sandia National Lab~mtories (Sandia) as part of the DBpm-tment of Energy'g (DOE) Accelerated Strategic Cornputin8 Initiative (ASCI) program. This s y a k r n ia named Red Storm and will be 3 distributed memory, massively parallel prrscaasm (Mpp) machine h i l t primarily out of c.cmumdity parts. The
more » ... mdity parts. The requirements presented here distill extensive architectural and design experience accumulated over a decade and a half of research, development and pmduction opmtiw of similar machines at Sandia. Red Storm will have an unu9ually high bandwidth, low latency interconnect, specially designed hardware and s & w m reliability features, a light weight kernel compute node *crating s y s t~m and the ability to rapidly switch major sections of the machine between classified and unclassified computing mvironments. Particular 31tmtim has been paid to architectural bdawcc in the design of Red Sturn, and it ia therefore expactad to achieve an atypically high fracti~n of it's peak s p e d of 41 TmOPS on real scientific mmputing applicatiosna. In addition, Red Storm is deaigned to be upgradeable to rnany times this initial peak capability while still retainiw appropriate balance in key d e s i g n dimmsims. Installation of the Rcd Storm computer system at Sandia's New Mexico site is plumed for 2004, and it is expected that the syatm will be operated fm a minimum of five years fobwing installatim, 3
doi:10.2172/918236 fatcat:347bmn533vcvfl5vx5637j6mma