The correct answer is "It depends".

In this case, what it depends on is several things.

What other workloads are on the drives?

What RAID5 (3+1 or 7+1)?

How many members in the meta? (I think maybe 4 from your post)

Are the meta members sharing the same physicals?

Is the workload sequential or random?

IO size? These look to be about 128KB IOs.

The throughput of a single disk is dependant on some of these parameters, so I can't answer that either, except to say "It depends"

As always when it comes to performances ..

Hi thanks for your answer.
Here are some precision, based on your question. In fact, I was more after the maximum performance our DMX may deliver, assuming the best scenario..

What other workloads are on the drives?
Let's assume none. The box is almost empty.

What RAID5 (3+1 or 7+1)?
7+1

How many members in the meta? (I think maybe 4 from your post)
META4

Are the meta members sharing the same physicals?
You mean physical drives? No. It is a Striped Meta.

Is the workload sequential or random?
It should be random. It is the DB volume of a MsSQL server. Logs are on another independant volume.

IO size? These look to be about 128KB IOs.
This is a good one. NTFS5 is default to 4KB cluster size, MsSQL2005 is default to 8k pages and the read ahead is 64k.

But we are not sure for the application itself, how it organise the info but we may assume something even bigger.


I hope it helps!
Thanks in advance.

Yes of course. I simply see it as a rule of thumb, for that particular scenario.

So you have 32 drives doing large block random read IO with no writes.

This should not be a Symm limited, given enough front end ports, but limited by the drives. We have seen about 200MB/sec with random large block IOs from a single 7RAID5 group, and you have 4 of those, so your 50MB/sec seems well within the comfort zone.

Remember the "It depends". If you add writes to the mix, or the IO size isn't large, the answer could be much different.

And one other point... That number was at the limit, without regard for reasonable response times. In most cases, we tell folks to target somewhere around 50% of maximum to have reasonable response times.

Yes you are absolutly right. For MsSQL, Microsoft also recommends to try not go over 50-60% of the IO/sec capacity of the drives, to keep the counters Average Disk Sec/Read and Write below 10-12 ms.