azurehpc/apps/ior
JonShelley eb6f01506d Cleaned up file 2020-02-26 22:22:46 +00:00
..
build_ior.sh updates to IOR for pipeline testing 2020-02-26 21:39:49 +00:00
ior.lsf updates for LSF and CentOS 7.7 2019-11-29 11:04:41 +01:00
ior.sh Cleaned up file 2020-02-26 22:22:46 +00:00
mdtest.sh fix load module issue 2019-09-27 12:46:41 +02:00
readme.md updates to IOR for pipeline testing 2020-02-26 21:39:49 +00:00

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>