Originally posted by: Newbie

Hello,

I was reading about disk IOPS and as per the info that I have:

RPM Drive Type IOPS

7200 SATA 80-90

10000 SATA 110-125

10000 FC/SAS 130-140

15000 FC/SAS 150-180

RAID Type Penalty

0 -- 1

10 -- 2

5 -- 4

6 -- 6

Please let me know if these numbers are correct.

Also, the formula to calculate RAID IOPS = (Total IOPS*Read percentage)+ (Total IOPS*Write Percentage*RAID Penalty).

So if this formula is true and the above Disk IO numbers and RAID penalties are correct, then to support an application requiring 500 IOPS, with 50% read and 50% write, I would need a RAID 5 config of 1250 IOPS. If I use 15K FC/SAS drives I would need a config of 6D+1P.

I am not considering the Cache on the subsystem, but a generic IO calculation from RAID perspective. Please let me know if I am missing some information or I have some incorrect information.

Thanks,

Yoko

Originally posted by: cris

Looks like you pulled this out of a manual.

The disks per IOPS look ok. it?s safer to use a range rather than an absolute value because factors like block size and locality of reference will affect the real world values. The "Penalty" section, which I assume you mean the number of backend operations per frontend operations, is a little out of whack. I'm not sure this is a problem, as long as you understand that a single front end I/O generally translates to many backend I/Os and you understand that with random workloads, there is always 2 parity operations per parity volume. IMO, unless you are running an extremely lean environment I wouldn't take it this far. Pools, gathers writes, rehits and caching in general make it almost impossible to understand what the real workload effects will be on the underlying spindles. I would just calculate the number of front end IOPS required for the application add 10%-20% (depending on how big the number are and factoring growth) and divide it by the disk IOPS to get the number of spindles. Realistically, unless it is a very specific application everything will be handled by a pool therefore you're dealing with a mixed workload environment.