Hi,
I've been wondering like forever why vSAN was performing rather poorly for me. I have a 4 node setup, with Xeon E3-1245 v6s, a single diskgroup per host (P4800x 375G/ P3600/P4510 2TBs).
Read performance was quite good, especially if I started multiple vMotions at the same time; I was able to reach several GB/s (moving to a nfs filer).
However moving VMs back to vSan was always kinda meh. Today I actually found a cause (though no explanation).
Moving a single VM (480G) from NFS to vSAN was performing quite good for a while until suddenly it became slow.
I checked the vSAN performance pages and found that basically the performance tanked when the cache disk was full and De-Staging was beginning.
Now the question is o/c, why an Intel P4510 NVME drive would not be able to ingest data at a fast enough rate to prevent the data transfer to come to down to some 100-200 MB/s ?
O/c Firmware is at the recommended level, inter Node connectivity is at 56GBit, cpu had at least some headroom (Peak usage was 12GHz, cpu has 4 cores x3,7GHz).
I have no idea of the inner workings of vSan, so maybe this has a very good explanation; I'd love to hear it in order to understand what the issue is.
Thanks
Hi,
Came here to say the same thing actually, EPYC 2 x 32c Platform, 2TB ram, P4800X with P4510 drives and when vmotioning to it bam it sucks:
Looks like when the write buffer is full it really starts to chug, I was thinking it might be the EPYC platform and NVME issues but to see you having the same issues I wonder if it's a P4510 bug?
Hi,
unfortunately this does not seem to be limited to the P4510's.
I have P3600s in my other nodes and they exhibit similar issues.
Is there anybody who reaches better DeStage performance with a single disk group?
If so what hw do you run?
Thanks