ESXi 6.5 datastorage low speed

[snowdog664]

Hi!

I have a Dell server with PERC H730 RAID card. CPU is Xeon 1220v6 and 32GB DDR4 RAM is present. There are 4 SATA3 2Tb HDDs configured as RAID-10 array. The driver is scsi-aacraid.

Also I have an old desktop PC with Adaptec 2408 RAID card. CPU is Core 2 Duo E8500 and 8GB DDR3 RAM is present. There are 4 SATA2 250 Gb HDDs configured as RAID-10 array. The driver is lsi_mr3.

All of the HDDs are desktop-class and 7200 RPM.

Both arrays have enabled write-back caching, PERC has 1Gb cache and BBU, Adaptec has 128Mb cache and no BBU.

Both RAID cards are updated with latest firmware, and the BIOSes of the mainboards are also latest version.

Both rigs have ESXi 6.5U2 installed and both ESXi's have Windows 2016 Server virtual machines running on them.

From that Windows VMs I ran speed tests.

This is the storage speeds of the old PC:

This is the storage speeds of the new Dell server:

Now comes the question.

Why the old desktop configuration from 2008 shows same or even better datastorage performance than a brand new Dell server equipment from 2018?

If it's due to a misconfiguration, how do I resolve this issue?

Thank you!

[snowdog664]

PERC has lsi_mr3 driver, and Adaptec has scsi-aacraid of course, my bad.

[daphnissov]

Well, just to point out a few things:

1. You're really comparing apples to oranges here as you have completely different everything on both systems. So because there are so many variables in play, it's almost impossible to point to any one.
2. The performance differences you're seeing are only lower on sequential read tests. All other tests (excluding seq write) are actually higher on the PERC.

[snowdog664]

I see your point, but don't you think that a modern RAID-10 array should show better performance than it does now? That is my question - where is the bottleneck.

[daphnissov]

Maybe? Possibly? It's difficult to know because you and I are both guessing here as we have no true comparative numbers. It'd be one thing if you had performance figures from bare metal and a second set from a virtualized environment, but you don't. Unless we have a fixed point of reference for comparison, there's not a whole lot we can do other than make general statements and guesses. So here's one: Ensure the VM from which you're testing this has a SCSI controller set to paravirtualized SCSI (PVSCSI) as this will allow for better throughput with lower latencies. Also check the available firmware on your RAID adapter and the drives for modern updates.