One

I was browsing through a publication today, it's aimed largely at big corporates and "decision makers" - so read " not technical people". Anyway it was talking about storage performance, a subject close to my heart!

Anyway the basis was the throughput that could be achieved with a new raid 5 array controller, amongst the claims were 200% faster than the previous card, and 380Mb/sec writes with raid 5 and 800Mb/sec read from a raid 5. Sounds good doesn't it? add a card and double performance? Well it's not quite as it seems, the product is a sata controller and the new card doubles the ports ( drives ) available - so 8 drives vs 4 drives yup about double performance ( this wasn't actually clearly stated ) - of course you've actually doubled the risk of failure of the array by increasing the disks in raid 5 (  each additional disk to a  raid 5 increases the risk of failure ). So throughput - again meaningless,  no specified data block size or sequential or random i/o.  SQL Server would make 8kb read/writes in worst case random i/o - the av scssi 10k drive supports about  120 i/o per second ( ref: sql server 2k perf tuning - ms press )  so my calc gives 7 x 120 x 8kb which is around 6Mb/sec - hmmm. - well it isn't even like that for writes as raid 5 cuts available i/o by 75% so we'd get a round 2 x 120 x 8kb which is just under 2Mb/sec.  The article figures actually indicate only a 50% ( ish ) write degredation vs reads for raid 5  ( oh this is supposed to be achieved with caching I think from the write up )  check out the perf tuning guide for write degredation for raid 5.

And no this wasn't an advert it was a case study. Hmmm - file under grumpy!!!

Published 09 October 2006 14:22 by GrumpyOldDBA
Filed under:

Comments

No Comments