Re: Optimal RAID configurations for relational databases

Hi

Malcolm Blackhall wrote in message <366F76CE.CC3CA1EF_at_pbnec.com>...
>Yes, you will take a performance hit using RAID 5 for two reasons. The
 first is
>the very nature of RAID 5. It maintains parity as part of each stripe to
 provide
>redundancy. Whenever you write to the device, the controller has to
 calculate
>parity and write it as well as the data. On the up side, you can get some
 read
>performance improvement under certain circumstances if the system is tuned
>properly.

100% agree.

[cut]
>There is another issue with the RAID 5 configuration you describe. The
 Smart 2DH
>controller is a two channel device. For optimal performance, you will put
 two
>drives on one channel, and three on the other channel. RAID 5 will protect
 you
>in the event of the failure of any one of these drives, however, if you
 lose one
>of the controller channels, you are out of business because you have lost
either

Another option would be to create single RAID-1 (using two disks on both channels) and single RAID-0 (using three disks, also on both channels, or on another controller). On RAID-1 put master device (msdb, msdblog etc. as well) plus log devices for your database. On RAID-0 keep nothing more than data devices for your database. Also configure your database NOT to use options "trunc. log on chkpt." nor "select into/bulkcopy" so you always can DUMP TRANSACTION . In worst case (one of RAID-0 disks failure or one channel failure) you can still DUMP TRANSACTION WITH NO_TRUNCATE to have most most up-to-date database backup. Of course this requires good backup policy, but you can have some performance benefit from using RAID-0 for your data devices. Especially if your database is write-intensive and RAID-0 is using own RAID controller.

[cut]
>Personnally, I would try to get another drive and configure the six drives
 as
>three RAID 1 devices. You get a little less storage and pay a little more,
 but
>get some peace of mind for it.

Sure, this is most secure.

--
Bronek Kozicki
--> remove __no__spam__ when replying