Re: RAID-5 and Oracle Performance

In article <6aqk1e$2l2_at_news.us.landisstaefa.com>, Jeff Shervey <jeff.shervey_at_us.landisstaefa.com_spamblock> wrote:
>Agreed... combinations of RAID-5 with a write-back Hardware RAID controller
>and RAID-1 work good. Of course, RAID 0+1 is the absolute best all around
>RAID configuration but it's expensive because every disk is mirrored besides
>being striped. No write penalty because there is no parity kept.
>Especially when restoring a 50GB database and a VP with a stopwatch is
>looking over your shoulder wondering when the production database will be
>ready. Also, no degradation when a single RAID-5 disk goes bad or is
>offline from the Raidset.
>We are using DEC Alpha 4100 based servers with HSZ-40 (32MB Cache) and
>HSZ-70 (128MB) RAID controllers.
>
>We just received the DEC HSZ-70 Enterprise Storage Array with about 54GB of
>9GB disk. (7 - 9GB disks) We were eating 2GB and 4GB disk up as if they
>were candy.
>
>/diskA-r5 - 54GB (9GB x 7 Disks) - Oracle data and indexes - tablespaceA -
>DatafileA
>
>My UNIX admin wants to throw our Oracle Manufacturing database on one single
>giant 54GB RAID-5 array. I don't really agree. Because Oracle data and
>indexes won't really be separated from disk drive heads. I would rather
>have 2 separate 27GB RAID-5 arrays. Example:
>
>/diskA-r5 - 27GB (9GB x 4 Disks) - Oracle data - tablespaceA - datafileA
>/diskB-r5 - 27GB (9GB x 4 Disks) - Oracle indexes - tablespaceB -
>datafileB
>
>Therefore when queries touch both the data and indexes - Physically separate
>drive heads on 2 separate arrays will not contend with one another. They
>will execute more in parallel.
>
>Does this make sense to anyone? What do you all think?

Doesn't your volume manager make it real simple to partition the array like that? Perhaps the real argument you should have is about separate controllers. Each controller is going to be busy figuring out what order to make efficient disk-head-movement requests, so if you can separate the index and data by controller, you remove a potential bottleneck, especially for full-table and full-index scans. You may want to reconfigure everything later anyway when you start monitoring the spread of requests across disks...

>
>jeff.sherveynospam_at_us.landisstaefa.com
>Take the nospam off my username to reply...
>
>Jeff Shervey
>Landis and Staefa - Buffalo Grove, IL
>Oracle/Ingres DBA - Oracle Financials
>
>
>
>

-- 
These opinions are my own and not necessarily those of Information Quest
jgarry@eiq.com                           http://www.informationquest.com
http://ourworld.compuserve.com/homepages/joel_garry
"See your DBA?"  I AM the @#%*& DBA!