I am considering implementing an iSCSI shared storage array using SATA SSDs instead of the 15k RPM SAS drives we normally purchase.
We normally use RAID 10 because of spindle contention with the random IO produced by virtualized workloads.
I was wondering if we could switch to RAID 5 or RAID 6 to have more usable space now that spindle contention is less of an issue. A question in my mind is how much overhead there is from the controller calculating parity. I am aware that this configuration will not allow TRIM to function.
Our current workloads are running on a Dell H800 with a 24 bay external enclosure.
Best Answer
Modification of single block will still require reading whole stripe from the disk and saving it again, or recalculating the parity block (depending on RAID implementation) and saving it to disk. Because SSDs have less problems performing random accesses it won't suffer as much performance degradation as similar array with HDDs but it won't be negligible.
The other problem you may run into is the performance of the controller itself. Cards with older chipsets can't process more than 400-500MiB/s with RAID. You definitely want to check for that.
In the end, only you can answer this question by benchmarking. Check how fast either configuration is running and use it.
I'd advise against RAID5/6, especially when it's a performance critical array.