Performance Sanity Check

[bg_IT]

Hello All,

I'm running some diskspd tests on our Starwind HA Cluster. I have 12 2.5 inch hard drives (10K - 600G) configured in Raid 10. It's a clustered Hyper-V environment. Testing from inside VM yields the following results for random 4k reads and writes.

I think these results are par for spindle drives but I'm not a guru. What does everyone else think, are these about right?

Command: diskspd -t4 -o8 -r -b4k -w100 -h -L -c20G

Result:

Write.png (68.81 KiB) Viewed 4532 times

Command: diskspd -t4 -o8 -r -b4k -w0 -h -L -c20G

Result:

Read.png (60.92 KiB) Viewed 4532 times

[Boris (staff)]

Turn your VHDX disks into fixed ones. I am sure you are using the dynamic one at the moment. Dynamic disks do not perform really well, especially in case of random write operations, due to the way how they process how they process them.

[bg_IT]

Sounds like it should be doing better, I just check it is a fixed disk. What else should I be looking at?

Disk Description:

Fixed VHDX.png (4.66 KiB) Viewed 4521 times

System Specs
Server 2016
2 x Dell R720
12 x Hitach Ultrastar HUC106060CSS600
Perc H710 Mini

[bg_IT]

The ISCSI target is using a NIC team, would that cause problems?

What would you expect to see for IOPS using starwind with a 12 10K disk Raid 10 array?

[Boris (staff)]

NIC teaming is not advised by any iSCSI vendor. Do not use it for iSCSI traffic. It will create additional overhead in transferring your data.

[bg_IT]

Thank you Boris, I've got a plan to give the servers a 10 Gbps server to server connection for ISCSI this weekend. I'll update this thread with new benchmarks when complete.

[Boris (staff)]

That will definitely be a worthy upgrade. Keep us updated.