1 x 9+1 600GB 15K SAS + 6+2 2TB SATA in Storage Pool
2 x 200GB SSD configured as Fast Cache
3 ESX hosts connected via 4GB FC (SAN is currently on 4GB FC as well)
SQL VM running Windows 2008 R2
4 vCPU
16GB RAM
40GB OS partition
3 x RDMs for User DBs, Logs and System DBs
Pathing Policy set to Round Robin
This server isn't in production yet but we ran SQLIO on it to benchmark it and we're fairly disappointed with the results. I would have thought with the EFDs the performance would be flying but apparently not. Any thoughts? Our DBA ran the SQLIO so I need to double check his configuration but here's the output from some of SQLIO.
Based on the 64 kB IO size I'd expect way less than 3500 IOps for each of the the flash devices, so I think what you're seeing isn't that bad, but I'd like to see what others think about this.
It wasn't formatted to 64kb sizes but I've now done this. I'll re-run the tests again but I was expecting higher performance especially with the fast cache.
you get pretty decent IOPS during your sequential test, flash drives are really great for the random workloads ..sequential workload they might be slightly better than SAS drives.
Interestingly I've been looking at our fast cache performance and according to the statistics its doing very little. for our main Storage Pool it's 0 for all the stats and for the RAID groups it's <20 for everything. I see more misses than hits.
This can mean that your locality is low and data cannot be reused a lot since it's not in cache. But the warm up is also something to consider. How long does your test run?
Normally FAST Cache is great for adding extra performance to SQL.
dynamox
11 Legend
•
20419 Posts
•
87439 Points
424
0
Posted November 3rd, 2012 15:00
windows drives were formatted with 64kb unit allocation size ?