Hi,
When running max iops workloads to 24 SDs with 64 threads each I noticed that outstanding qdepth reported by vdbench was quite less than max value. It was ~900 on average while max is expected to be 24x64=1536.
So there were some threads in 'transit' state, about 26 per SD. I thought it had to be close to max except a few threads per SD.
Vdbench was configured to generate both compressible (50%) and dedupable (ratio=2; dedupsets=N which is 5% of a LUN size, not all the test areas) data. Clients are linux. Reproduced both on VMs and physical ones.
Is it something expected? Can you please advise what needs to be checked if not.
Thanks
Rustem