Starwind with Huawei Tecal ES3000 low performance

[garciello@maiora.com]

We have Windows 2012 cluster with 2 nodes with starwind 6.0.5437 (2 node with 4Tb license).
Storage array: 24 sas 15K + 1 Huawei Tecal ES3000 2.4TB SSD per node

We tested the performance Huawei Tecal with Iometer (4Kb 100% read; 100% random):
1) Huawei Tecal no cluster without StarWind, performance is good

Image1.PNG (37.28 KiB) Viewed 4891 times

2) Huawei Tecal without cluster and with StarWind, performance is half that of the first test

Image2.PNG (36.71 KiB) Viewed 4891 times

3) Huawei Tecal with clusters and with StarWind, performance is reduced by 1/3 compared to the second

Image3.PNG (32.82 KiB) Viewed 4891 times

In the end, the performance with Huawei Tecal are the same as with the disk array sas 15k.

Any suggestion?

[nohope]

I have serious doubts regarding the amount of IOPS you got without SW and cluster. Don't you mind that? Really poor performance for such device as, normally, PCIe SSDs should break 300k IOPS threshold on random reads. I really advise you to check your SSD configuration. Double check where it's being installed - PCIe slot should be x8 v2.0 in your case. Updating drivers/firmware is also a good idea.

[garciello@maiora.com]

Yes I know,
but however perfoming same test with starwind, I get half perfomance.

[anton (staff)]

Version of the software you're using is very old and is not even supported already.

I'd recommend to get in touch with engineers they will check how much we can squeeze from your setup using up-to-date V8 builds and you decide worth it discounted upgrading or not.

Yes I know,
but however perfoming same test with starwind, I get half perfomance.

[garciello@maiora.com]

I have already requested an offer for upgrade two days ago...
I'm still waiting

[Max (staff)]

Hi!
As far as I see from our CRM you should've received a quote for the update yesterday.
Please PM me if that's not the case and I'll take from there.
In the meantime +1 to answers above - 6K IOPS means something is seriously broken.
I'd recommend to get the latest version as well as one of our engineers remoted in for investigation and troubleshooting.

[garciello@maiora.com]

Hi,
i've upgraded starwind to version 8.

I repeated tests with Fio (Flexible IO) benchmark software (100% sequential read with 4k cluster):
1) Testing directly on Tecal 3000 I got 210.000 iops
2) Testing on Starwind disk single node (no HA) I got 30.000 iops

Why so much performance difference?
What other tests can I do?
Thanks

[anton (staff)]

Longer I/O path. Incorrect cache policy. To few LUNs. Incorrect mapping of StarWind vLUNs -> hardware LUNs.

There's not much we can do with 1) but for 2) and 3) and 4) kick support guys so they'd jump in on your test bed and do a proper setup for you getting all possible IOPS from your config. That's doable at least with aggregated model (many storage consumers + many storage providers).

Hi,
i've upgraded starwind to version 8.

I repeated tests with Fio (Flexible IO) benchmark software (100% sequential read with 4k cluster):
1) Testing directly on Tecal 3000 I got 210.000 iops
2) Testing on Starwind disk single node (no HA) I got 30.000 iops

Why so much performance difference?
What other tests can I do?
Thanks

[Vladislav (Staff)]

Quick community update:

We are already working on this case.