Hey folks,
I'm speccing a SAN array (EMC VNXe 3200) for a brand new VMWare implementation running six Windows 2012 guest VMs. The client has pushed back on the configuration I specced, asking why not use a configuration with slower drives but more storage space for the money. Here are the differences:
I specced | client Googled |
---|
twelve 600GB 2.5” SAS drives at 10,000 RPM | 12 x 4TB 3.5” NL-SAS drives at 7,200 RPM |
total RAID 6 storage = ~5.3TB | total RAID 10 storage = ~22TB |
$21,500 | $21,000 |
Please note (a) difference in RAID levels possible due to differences in drive capacities and (b) the 3.5” drives are NL-SAS but the 2.5” drives are pure SAS.
This prompted me to go to sleep that night thinking, “Gee, I was wondering when the day would come when I would finally do that much research on IOPS and RAID workloads.”
Based on my completely novice understanding of IOPS and RAID workloads, using EMC’s published drive W/R and latency times at http://tinyurl.com/cq577xe and based on an anticipated six Windows 2012 Server VMs (running standard office productivity-type workloads (no virtual desktops)), I calculated average IOPS under various W/R workloads and different RAID levels, and concluded that the IOPS numbers between the two configurations are roughly the same.
The relative parity in IOPS numbers is due to the fact that, despite the lower RPM of the NL-SATA drives, RAID 10 has a much lower write penalty than RAID 6. And actually, the RAID 10 configuration detailed above could reserve a drive as a hot spare and would still provide wayyyy more storage space than the client needs.
In case it matters for this discussion, the figures I arrived at for the entire array are:
W/R 80/20% => 398 IOPS
W/R 70/30% => 572 IOPS
W/R 60/40% => 746 IOPS
Assuming I didn’t munge the calculations (which of course is likely :-), and aside from the fact that NL-SAS drives have a higher MTBF than SAS drives, what else should I consider before telling the client, “Yeah okay, your suggested SAN array is better for the price”?
<whew>
Thanks!
noam