eb6f01506d | ||
---|---|---|
.. | ||
build_ior.sh | ||
ior.lsf | ||
ior.sh | ||
mdtest.sh | ||
readme.md |
readme.md
IOR and mdtest
IOR can be built on any kind of VM SKU as long as it uses the CentOS-HPC 7.6 image. See one of the examples for building a Cluster with HB or HC skus and PBS. (e.g. simple_hpc_pbs)
Pre-requisites
The pre-requisites for running IOR are those :
- jq
- mpich-3.3 for HB/HC
- mpich-3.2-devel for others skus
The build script will installed the missing component by default
Log-on to the headnode
azhpc-connect headnode
cd /apps
git clone https://github.com/Azure/azurehpc.git
Build IOR
Build IOR/mdtest from the build script. This will by default install ior and mdtest into /apps/ior and create a modulefile. You can override the installation path by providing it as a parameter on the build_ior.sh
script
For PBS :
qsub /apps/azurehpc/apps/ior/build_ior.sh
For LSF :
bsub -q <queue> -o %J.log -e %J.err "bash /apps/azurehpc/apps/ior/build_ior.sh"
Run IOR and MDTEST
Now submit and run (e.g on HB):
For PBS:
qsub -l select=2:ncpus=60:mpiprocs=15 -v FILESYSTEM=<DIR> /apps/azurehpc/apps/ior/ior.sh
For LSF on 2 nodes, using 8 process per node :
bsub -q <queue> -R "span[ptile=8]" -n 16 -o %J.log -e %J.err -env "all, FILESYSTEM=<DIR" "bash /apps/azurehpc/apps/ior/ior.sh <DIR>"
Note: this will run on 2 node and 15 processes per node.
The ior.sh
script runs a throughput (N-N and N-1) and IOPS test.
A metadata I/O benchmark test can be run using the mdtest.sh
script.
qsub -l select=2:ncpus=60:mpiprocs=15 /apps/azurehpc/apps/ior/mdtest.sh <DIR>