IOR filesystem benchmark Scheduling frequency 6h , nodes=16:ppn=7:walltime=00:10:00
This is a small benchmark that matches the one on jyc: #PBS -lnodes=16:ppn=7,walltime=00:10:00 , using a stripe8/ directory created via "lfs setstripe -c 8 stripe8" . Performance should meet or exceed jyc's performance on average (jyc is typically dedicated while bluewaters is highly shared). The idea is to have a baseline of performance without being overly intrusive on the batch system or filesystem.
bwjenkins has been added to the acl for ~arnold/ior and as a secondary objective, the ior binary is rebuilt for each trial -- testing the compilers and environment on blue waters.
Tests: automake, autoconf, and default build environment, Lustre striping and throughput for a small job, scratch
This is a small benchmark that matches the one on jyc: #PBS -lnodes=16:ppn=7,walltime=00:10:00 , using a stripe8/ directory created via "lfs setstripe -c 8 stripe8" . Performance should meet or exceed jyc's performance on average (jyc is typically dedicated while bluewaters is highly shared). The idea is to have a baseline of performance without being overly intrusive on the batch system or filesystem.
bwjenkins has been added to the acl for ~arnold/ior and as a secondary objective, the ior binary is rebuilt for each trial -- testing the compilers and environment on blue waters.
Tests: automake, autoconf, and default build environment, Lustre striping and throughput for a small job, scratch