Disk Mark Questions

[mike@arnoldgauge.com]

Hi all, I have some questions on interpreting some disk mark results. Key parts of the testbed are:

• Dell R720DX server
• 128GB Ram, 2x E5-2690 CPU
• LSI 9207-8i SAS/SATA Card - Flashed to the latest IR firmware
• 2x Samsung 860 EVO 1TB SSD Drives
• Windows 2016 Standard - Not virtualized, installed on the SSDs

My goal is to determine the performance penalty for mirroring the SSD drives, using hardware mirroring with the LSI Card.

============

I've included pictures below, but to summarize:

In unmirrored mode:

• Disk Mark = 4627
• Disk Seq Read = 453
• Disk Seq Write = 403
• Disk Random Seek + RW = 422

In mirrored mode:

• Disk Mark = 2565
• Disk Seq Read = 455
• Disk Seq Write = 203
• Disk Random Seek + RW = 50

======

Observations:

• Disk Mark in unmirrored mode matches other reports from other systems. It tells me this system is performing normally in this mode.
• Sequential read performance is unaffected by the mirroring. That makes sense.

Questions

• Sequential write speed is 50% when mirroring. Is that expected when using hardware mirroring? I would have thought the card could drive more writes in parallel to the 2 drives. Am I missing a tuning parameter somewhere?
• What is the Disk Random Seek + RW telling me? Going from 422 to 50 doesn't make sense

[David (PassMark)]

If speed is really an issue you might be better off with a newer M2 SSD setup. You would get at least 4x the performance.

I would have also thought the RAID controller would to parallel writes. With the result being the slowest of the two drives. I am not really familiar enough with the LSI 9207-8i to offer any tuning advice. Did you look at the caching policy (Writeback vs WriteThrough)

Try Microsoft's dynamic disks (software RAID) with direct SATA connections to the motherboard. You might get a better result.

There is also the Advanced disk test in PerformanceTest if you want to experiment with different scenarios.