Connecting external storage will work on SImplivity?

[AnuragJha]

If the VMs are using SimpliVity NFS storage the load will be on OVC.

If you like to connect NAS / SAN storage, best option is to use a compute node which will work both with SimpliVity and NAS / SAN storage and you have the option of moving the data between both the datastores. This way there is no load on the OVCs

[Mike__W]

Steez, have you managed to figure out a solution for the high IOPS load? We've had a ticket logged with HPE to look ast the OVC but my gut feeling is our DB servers are just too demanding and causing some issues. We have added an 8th node to each of our simplivity clusters so I'm hoping the load will be balanced out a little more.

[steez]

@Mike__W I think you have the correct feeling, we also involved HPE and VMWare support, which from both sides told us that the hardware and VM are okay, that the issue is inside the VM, odd because in the old environment we had no such issues.

And unfortunately no we have find a solution. We have 2 issues. For some reason SAP and other DB like programs take longer to finish tasks (whereas in the old environment it was no issue (even when running on HDDs)). The issue really isn't the CPU here.

The second issue is high OVC load when data are being copied (specifically db from one VM drive to another). When our SVT hits 100% we see - 400MB/s troughput, latency goes up to 30-70ms, IOPS peak about 8k - that might be the max that our SVT can do.

Let me know about your results, however I think that adding node might not help (only if the current deployment is already congested with IO).

[Mike__W]

Our expansion was mainly because we're expanding the compute and nodes were starting to complain about running out of capacity so it wasn't just because of these issues.

We're seeing the same behaviour, roughly 400MB/s with the OVC sat at 100% and then we start having IO performance issues on other VMs using the same OVC, my gut feeling is that the OVCs struggle with very high amounts of read IO, especially since they are performing compression and deduplication at the same time so we're potentially looking at moving the high IO workloads to a non-simplivity datastore, not ideal we we lose a lot of features but this might be our only choice if we want to not suffer performance issues.

[steez]

@Mike__W This is very interesting, also somewhat confirms my suspicion about the what is the maximum that Simplivity can do in these DB conditions. Thank you for the follow up, good to know that im not alone.

I would wish for HPE to comment on the matter as Support didn't mention that the Simplivity might be maxing out on performance, just that it is normal for OVC to be 100% sometimes.