This repo will run vdbench against a storage target using an Azure CycleCloud compute cluster

Accept terms and download vdbench50407.zip from Oracle

NOTE: The vdbench User Guide is also available on the Oracle download page

SSH to your cluster scheduler node (aka head node) and run the following commands:

sudo mkdir -p /shared/apps/vdbench
sudo chmod 777 /shared/apps/vdbench
git clone https://github.com/themorey/cyclecloud-vdbench.git /shared/apps/vdbench
#NOTE:  install git if needed:  sudo yum install -y git

Copy (ie. scp) vdbench50407.zip to your scheduler node and unzip it into /shared/apps/vdbench...ie:

sudo unzip vdbench50407.zip -d /shared/apps/vdbench/

Run the following commands to unzip the filesystem tests into the appropriate directory:

sudo mkdir -p /shared/apps/vdbench/benchmarks/vdbench/
sudo unzip /shared/apps/vdbench/filesys.zip -d /shared/apps/vdbench/benchmarks/vdbench/
sudo chmod -R +x /shared/apps/vdbench

The script named vdbench-sbatch.sh will schedule this job to run across 24 execute (aka worker) nodes in the Slurm cluster. You can modify the partition and # of compute nodes requested by modifyig the following script lines:

# change partition from default to htc
sudo sed -i 's/##SBATCH --partition=default/#SBATCH --partition=htc/g' /shared/apps/vdbench/vdbench-sbatch.sh

# change # of execute nodes from 24 to 12
sudo sed -i 's/#SBATCH -N 24/#SBATCH -N 12/g' /shared/apps/vdbench/vdbench-sbatch.sh

Run the following command as follows to submit the job to Slurm:

NOTE: Run with a user account that has passwordless sudo as the scripts need to mkdir, install openjdk, etc

sbatch /shared/apps/vdbench/vdbench-sbatch.sh /data

In the above submission example the argument /data is the mountpoint to benchmark with vdbench. This could be an HPC Cache export, xNFS on Blob, Azure Files NFS, Azure NetApp Files, etc.

This will write test data to /data/vdbench and at the end write output/error files to /data/vdbench-output (assuming your test mountpoint is /data)

NOTE: The vdbench-local.sh script uses the local ephemeral disk on the compute VM. If usiing a VM size without an ephemeral disk you will need to modify the vdbench-local.sh script to a supported path for writing the output file