MSA 2040 Controller performance bottleneck

[steez]

Hello Everyone,

We have a MSA 2040 storage array with 20 1.8TB SFF SAS drives, and 4x400GB Read-cache SSDs for each controller respectively. 10 disks each pool and volume with RAID6. We would like to add an D2700 enclousure populated with 12 LFF SSDs. 

Average IOPS that performance monitor lists for Volumes are about 300-2000 depending on hot hours. Controller CPU load is about 25-30%

The question is - how much IOPS can MSA 2040 controllers process before they become bottlencks? Couldnt really find such information online, only some various test numbers. Can try to provide additional numbers, screenshots, im new in storage world.

[HPSDMike]

I'm not aware of published maximums from a controller standpoint. However, looking at the product quick specs, you'll find a multitude of performance test numbers noted in the tables. These were actual performance tests conducted, with the exact configuration noted. From my perspective, I would think you could look at the "all SSD" tests as being fairly closely related to what the controller limits would be.

https://h20195.www2.hpe.com/v2/getdocument.aspx?docname=c04123144

Also, while it is not required, you will be able to get the most performance potential out of the array by having all your SSDs in the base shelf with the controllers. This allows access to the SSDs without traversing the SAS interconnects.

A D2700 is a SFF shelf, you mention LFF drives. Also, the D2700 has recently been replaced with the MSA 2050 SFF Expansion shelf. You can use this on your MSA 2040; just make sure the controller firmware is at GL225R003.

[steez]

Hi Mike,

Thank you for the swift reply. Just to clear things out - basically if Controller Performance tab lists that controllers ar pushing through 2-3K IOPS each, that means that additional expansion should be no problem and those controllers are doing just fine?

And another thing, in tests mentioned in the document, it says that with HDDs it can achieve 13K IOPS, but what about when mixed configurations comes in place when I have for example half of my drives SSD and the other half HDD? Will the number be the same due slowest link in the deployment?

BR,

[HPSDMike · Accepted solution]

Yes, for the sake of this discussion, I would consider the 122,000 (8k) read IOPS (noted in the table) an indication of maximum tested controller IOPS. The measurement of a IOPS can be hard to interpret between different users. Your workload may be different from mine. The amount of IOPS you can drive is dependent on block size, read/write percentages, RAID type, host connectivity type, host connectivity bandwidth, how sequential/random the workload is, and how much latency you can tolerate.

So, in any case, you are listing your system as pushing about 3,000 IOPS per controller so we obviously have a lot of headroom between that, and the 61,000 IOPS per controller, noted in quickspecs max read test. I would expect you can continue to drive more performance out of this system as you add resources.

You can mix SSD, 15/10k, and HDD in an MSA 2040. With linear vdisks your max IO would be determined by the quantity of disks in that vdisk (with help from cache to improve those numbers).

If you are mixing disk types in a pool (virtual storage), and have the performance tiering license, then your worst case scenario of the performance of the slowest disks in that pool and your best case would be the performance of the fastest disks in the pool (plus cache). Your actual performance will depend where the data lives when it is called (where the MSA placed it based on tiering algorithms) plus whatever benefits caching is providing to you.

I've seen a 1GB iSCSI MSA 2040 push over 25,000 IOPS with only 12 nearline drives in them. Now, the latency was about 25ms and the test workload was fairly cache friendly, but it could do it. I've also seen that same MSA 2040 push well over 10,000 IOPS with only 2 SSD's and a fairly intensive I/O test going on that wasn't really cache friendly. Latency in that case was under 3ms.

So, your mileage will vary. However, it seems safe to say that you have plenty of room to grow before the controller becomes your bottleneck.