LLNL I/O Testbed Analysis of IBM M80 (P660 6M1)

Fitzgerald, K ; Daveler, J ; Heer, T ; Gleicher, M

Lawrence Livermore National Laboratory

关键词: Capital;    Lawrence Livermore National Laboratory;    Storage;    Computer Architecture;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;   

DOI  :  10.2172/15013326
RP-ID  :  UCRL-ID-148122
RP-ID  :  W-7405-ENG-48
RP-ID  :  15013326

美国|英语

来源: UNT Digital Library

【 摘 要 】

The HPSS hierarchical storage system currently uses ''mover'' nodes to interface client processes to the hardware devices. As archival bandwidth requirements increase, the number of mover nodes required to provide the necessary bandwidth has become a significant factor both in terms of capital expense and administrative overhead. The LLNL data storage group would therefore like to select the most powerful yet cost effective IBM architecture currently available for use as future HPSS movers nodes. The LLNL storage group currently uses IBM ''winterhawk2'' SP nodes as HPSS mover nodes. Several IBM architectures including ''nighthawk'' SP nodes and several ''P'' series IBM architectures were considered and the ''M80'' system appeared to be the most promising based upon its high memory bandwidth and capability to attach multiple ''RIO'' drawers offering 64bit 66MHz PCI buses.