Performance Sizing for Large Sequential Operation

We do know that per 10K FC disk will yield ~100 IOPS. If a customer requires 1000 IOPS, we will just size it with 10 data disks, but that's for small random ops. What if a user requires 100MB/s of large sequential operation? Is there a rule to say how much throughput will a FC or SATA disk generate?

Re: Performance Sizing for Large Sequential Operation


Welcome to the community

Each IOP is equal to 4Kb or 4.096kB/s of throughput and every disk will have a greater latency as the total IOPS increase.  If the data is random the IOPS will reduce as the disk read has to spend more time seeking for the data and as 10k spins slower than 15k, the seeks are longer.  Sequentail reads will be much faster (maybe 3x) due to caching/streaming technology.

If you have access to the data run the statit command to see what the disks are doing.

xfers x chain x 4.096 = throughput

this will enable you to better 'guess' when correctly sizing your aggregate.

Best guess for disk performance

SATA 500Gb 7.5k  = 55 IOPS

300 GB 10k = 120 IOPS

300 Gb 15k FC/SAS = 220 IOPS

These figures can and do change dependant on your data patten.