I am setting up an Exchange "Database". I set it up as Striped R5 Metaluns across 4 RAID Groups, each RAID Group on a seperate DAE on seperate busses (as I was "taught".). Unfortunately the DBAs want it on 25 disks, not 20 disks, according to their "IOPS requirements". So, now instead of spanning 4 busses equally, I'm "duplicating" a bus:
Old: 0_0, 1_0, 2_0, 3_0
New: 0_0, 1_0, 2_0, 3_0, 2_2
Which one of these layouts do you think will give better performance?
Great job getting IOPS requirements! Not everyone even gets that much info.
If your calculations say you need more drives, then that will usually be your best option.
Unless you are worried about extreme availability, you will most often benefit from the extra spindles.
Probably the reason why you were "taught" to go across only 4 loops was for availability.
If you striped your RAID 5 groups across all four loops, you would be better protected, In the unlikely event that you lost a DAE.
If you lost DAE 2_0 in your example, then you would likely suffer an outage due to the fact 2_2 would likely be affected.
(It's pretty rare that you would loose the whole DAE, but some architects design their systems this way.
Some also do it for performance, but that reason alone wouldn't stop you from using 2_2...)
You are likely never to notice the difference between the two. You've already done as much optimatization as is practical given your resources.
Any further optimizations would likely involve cache configuration.
The reason that I create Striped MetaLUNs across all 4 busses is so that parallel reads/writes are distributed across all 4 busses at the same time. If you stripe across 5 DAEs on 4 busses you are going to have 2 I/O requests at the same time on the same bus. Wouldn't it be faster to just have 4 I/O requests, one on each bus, at the same time - so they don't get in each other's way?