RAID 1 NVMe vs RAID 1 15K SAS

[rsumperl]

Hello everyone,

I’d like to hear some experiences with NVMe drives in RAID 1 vs 15K SAS drives in RAID 1. I am not seeing much of a performance difference between the drives. As an fyi, the RAIDs were created on a Dell H755. I had assumed the NVMe drives would be way faster and they’re not.

Thanks
Ray

 

[mattventura]

Try doing software RAID with the NVMe drives. The H755 is probably the bottleneck.

Edit: I should clarify. I mean you should try bypassing the H755 altogether, not just configuring them as JBOD drives.

 

[rsumperl]

The motherboard has oculink connections on it and I do have the configuration plug. I will try direct attachment to the board.
The RAID card does seem to be the bottleneck as I tried RAID 1 with SAS SSD drives and they were the clear winner. Mind you these test were all done by “gut feel”, but the SAS SSD drives were fast. I now know where my exchange database will reside.

 

[XeonSam]

There's no way 15K sas spinners or sas ssd's can perform as well as nvme drives. Try using a benchmark tool because the gut feel is not so accurate. You'll see that anything that demands extended high IOPs will do much better on nvme drives. The H755 has a butt load of nvcache, make sure the data set is larger than the cache by atleast double and run your benchmarks.

I tried this with H730 raid cards on 7.2K drives with a dataset equal to the cache and the performance was equal to an SSD. This was 4 drives though, but the cache really does help with performance on dell servers. Even more so with HP servers.

 

[rsumperl]

Thank you for the advice, how do you configure the data-set?
Is this a CLI command?

Thanks
Ray

 

[XeonSam]

This really depends on what tool you're using. It's just the test size - usually it's defaulted to something small like 1GB. If you're not used to benchmarking tools, use something simple like crystal disk. It's not too accurate for server type workloads but the other tools are much more complicated to use.