On Thursday 29 May 2008 05:33:44 pm Dean wrote: > RAID5 performance is *highly* dependent upon how fast your controller > is, and how much offloading it does to your CPU. This is because it > requires parity information to be calculated for every write. > > Most INTEL/AMD/PROMISE/HIGHPOINT controllers are to a varying degree > fakeraid, and are very poor at raid5. Especially those integrated into > mainboards. > > RAID 5 doesn't *have* to be killer slow, but the pricepoint most people > are willing to pay means they get poor raid5 performance. > Even with RAID controllers in the four digit price range RAID 5 write performance is far slower than an equivilent controller using RAID 10. It's a horrific choice for a database that is going to see any sort of write load. Not only do you have to calculate parity, but you have to seek every head in the array to do a write. In practice on a heavily loaded database server seek times on the drives dictate throughput. I have database servers with RAID 10 arrays of 15k U320 drives that never manage to do 20Megs/sec of throughtput, even though they are completely I/O bound. While a sequential read or write can be nearly 15 times that. -- Thanks, Josh Paetzel PGP: 8A48 EF36 5E9F 4EDA 5A8C 11B4 26F9 01F1 27AF AECB -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 195 bytes Desc: This is a digitally signed message part. Url : http://mailman.mn-linux.org/pipermail/tclug-list/attachments/20080530/457e3348/attachment.pgp