Enhance fio benchmark to measure latency, IOPs and bandwidth #551
Conversation
and throughput. There is an increase in disk space needed to run now, from 5G to 16G.
|
Rebased to head |
| ramp_time=5s | ||
| runtime=120s | ||
| ioengine=sync | ||
| # check if .StorageDir is a directory |
There was a problem hiding this comment.
The checks for existing, writeable directory with enough space to run the benchmark were added because these issues were run into by users in the past. Without these, diagnosis of failure to run was difficult.
| fi | ||
| # create temporary directory for fio test | ||
| test_dir=$(mktemp -d --tmpdir="{{.StorageDir}}") | ||
| sync |
There was a problem hiding this comment.
sync and dropping caches isn't strictly necessary, but are good practice when running storage benchmarks
|
Didn't try it yet, but assume this will fail on ARM targets because libaio support is missing. |
1aa168e should cross compile libaio and enable fio cross-compile to statically link libaio, it requires a newer unreleased fio tho |
insufficient diskspace log partial storage benchmark output if unable to find fio output
bandwidth test which uses the most diskspace Add back in sync and dropping vm cache
Completed testing, appears to be working fine on C4A (Arm neoverse v2) |
|
The build is failing. Run |
|
Surprisingly, I was able to update the branch with the small change needed to get past the quality checkers. |
harp-intel
left a comment
harp-intel
left a comment
There was a problem hiding this comment.
Thank you @echiugoog
Great improvement to this benchmark.
2 participants
FIO benchmark updated to use a job file and output in JSON