作者: Charng-Da Lu , James Browne , Robert L. DeLeon , John Hammond , William Barth
关键词:
摘要: This paper presents a methodology for comprehensive job level resource use measurement and analysis applications of the analyses to planning HPC systems case study application XSEDE Ranger Lonestar4 at University Texas. The steps in are: System-wide collection performance statistics node levels, mapping storage resultant job-wise data relational database which eases further implementation transformation formats required by specific statistical analytical algorithms. Analyses can be carried out different levels granularity: job, user, or system-wide basis. Measurements are based on novel lightweight job-centric tool "TACC_Stats" [1], gathers set metrics all compute nodes. tools will an extension XDMoD project [2] community. also reports preliminary results from measured Texas Advanced Computing Center's supercomputers. studies presented indicate detailed information that available resources when TACC_Stats is deployed throughout system. applied any system runs tool.