Server Fault is a question and answer site for system and network administrators. It only takes a minute to sign up.
Sign up to join this community
Your bandwidth/throughput should go up with raid 5, but you will take a bit of hit in latency after you over run the cache, since you need to calculate the parity.
Has anybody done a latency test using 1 HD vs 3 HD in raid 5? I am very interested to see if SSD makes any difference. (My prediction is SSD would expose this latency, since it would now be in the same order of magnitude)
It depends on the type of transfer. I don't have any exact figures (I have planned for a while to do a mini study for my own reference, since discovering Linux's read performance on RAID1 arrays (with Debian/Etch's stock kernel at the time - I gather an improvement has been made recently) wasn't quite what I was expecting, but haven't got around to finding the time...), but for read performance (both random and sequential) you should expect a healthy 3-drive RAID5 array to respond like a 2-drive RAID0 array.
For writes you will hit the well known problem where-by a write becomes a read and two writes (read the other block of the block+block+parity set, recalculate parity, then write the new block and new parity block). Exactly how much this affects your system depends on the write load and pattern. If you are running a database that sees many transactions you want RAID10 not RAID5 (to avoid the write performance hit) but for a volume that sees few writes but relatively many reads RAID5 will improve performance over a single drive, not degrade it.
As the drives are distinct I would not expect the intelligence of the SSD to m,ake much difference (they are different drives so write combining will not help) but I would not expect to see a massive difference in relative performance for sequential writes. If your spinning disk based array takes 2.0 times as long as a single spinning disk to run a sequential write then an SSD based array will take 2.0 times as long as a single SSD. For random writes I would expect to see the SSDs perform much better relative to a single SSD then an array if disks does relative to a single disk because of lower latencies due to not needing to move heads and wait for disks to be in the right spot in their spin cycles.
Well, no, but I can say that with a good battery backed RAID controller like the Dell Perc6/i the RAID5 performance is very good. My somewhat vague recollection of the last time I tested this is that with the Perc a 4 disk RAID5 wasn't far behind a 3 disk RAID0. The controller is so efficient at doing the parity calculations that it adds negligable overhead.
On my Poweredge 2950 Hyper-V servers I use a 6 disk RAID5 and it's the fastest array I've ever seen outside of specialist SAN kit. Ten years ago I would have said use RAID10, but these days I have no hesitation about recommending RAID5, or better still RAID6 if you have room for the extra disk.
JR
