Skip to content

Latest commit

 

History

History

nfs-io-performance

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 

NFS Server Provisioner IO Performance

Determine performance delta between using Multus and Calico for NFS client/server traffic. Many other permutations of the final configuration were run but this is a sample test.

Configuration

Using benk-v0.0.1 to scale a mixed workload.

benkWaitForPVCs=1
pvcVolumeSize=512Gi
workloadRuntime=60
workloadWorkingSetSize=512M
workloadThreads=16
pvcPersistPVC=1
pvcAccessMode=ReadWriteMany
workloadDeleteFiles=1
# Change these
workloadPattern=randrw
workloadReadPercentage=80
workloadBlockSize=8k
workloadReplicas=%workload%

Start workloadReplicas at 4 and double each iteration up to 512.

export workload=4
for i in {1..8}; do
  cp -a kustomize/overlays/default kustomize/overlays/example-${i}
  sed -e "s/%workload%/${workload}/g" examples/nfs-io-performance/config.env > kustomize/overlays/example-${i}/config.env
  workload=$((2 * ${workload}))
done
./sequencer.sh example-

Report

Create the output.

./src/benk/outputter.py -a examples/nfs-io-performance/a.log -b examples/nfs-io-performance/b.log -t examples/nfs-io-performance/template.tsv.j2

The output is in TSV and is meant to pasted into a spreadsheet.

Example Rendering

Multus vs Calico