Re: Q: Is RAID/striping good with Oracle???

Do you have any concern for failure of the cache memory in the disk controller? It would seem you're trading off failure of the disk controller vs performance. In that sense why bother with the RAID 5?

 Albert W. Dorrington (awdorrin_at_ictest.delcoelect.com) wrote:

: In article <4vv2n9$keg_at_larch.cc.swarthmore.edu>, stauffer_at_crabapple.cc.swarthmore.edu (R Glenn Stauffer) writes:
: |> In article <4vuv3h$36h_at_lal.interserv.com>,
: |> Bill Meskill <meskillw_at_ncr.disa.mil> wrote:
: |> >I'm running RAID 5 on my HP/UX prod server and wondering
: |> >why you said to stay away from it... Can you elab your
: |> >experience/finding?
: |> >
: |> >Thanks
: |> >Bill Meskill
: |> >meskillw_at_ncr.disa.mil
: |>
: |> Common wisdom has it that there is a significant write penaltly with Raid5.
: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 

: Make that 'Common Misconception has it that there is a significant write
: penalty with RAID 5.' It all depends on how how your controller is configured.
: also, depending on how much writes versus reads you will see the overall
: performance could be _FASTER_ due to the RAID controller being able to
: pull data off the disks in parallel.

: |> I've also heard some suggest that you will lose the entire raid volume if you
: |> lose two drives - I'm not sure that is any more likely than losing both drives
: |> in a mirrored set.
 

: Depending on how you have your RAID controller configured, this may or may
: not be true. In my case I have RAID controllers which are configured
: with 7 2.1 GB disks (14.7 GB total) Six disks are striped across each other
: and one disk is left for a Hot Swap replacement.
 

: If one disk fails, the hot swappable disk cuts over and begins to sync
: itself up to replace the failed disk. (This takes 45 minutes for my
: particular controller.) After this time period, everything is fine again,
: and you can lose one more disk, without losing data and without putting
: any additional stress on your day. Without RAID - after just the one disk
: crash you'd have a few hours of work to do to recover that particular table
: or tables on that disk.
 

: |>
: |> If you run RAID5 on a transaction system without a caching controller with a
: |> fairly large cache, you will see a performance hit. If your system is a read
: |> intensive DSS, that shouldn't be as true (of course, Oracle is still doing
: |> writes for sorts, etc). I read something when I was configuring our server
: |> that you need something on the order of a 40mb write cache on the controller
: |> to eliminate most of the RAID5 write penalty. Of course, then you have to
: |> weigh the cost of more drives (RAID0+1) vs more costly controller hardware
: |> for Raid5.
: |>
 

: The use of RAID 5 controllers for a database system depends
: totally on what type of transactions that your database will be
: encountering.
 

: In my particular case, I have a Oracle database on a SPARCServer 20.
: The database files are distributed over two RAID controllers of 10.2GB
: each (total of 20.4GB of storage). We have many small write transactions
: and many large read transactions. (We average maybe 75% Reads and have
: a few thousand inserts per day.)
 

: We have not seen or had _any_ performance hits with this
: array. The SCSI bus is Fast, Wide, Single-Ended, and there is 16MB of
: cache per controller. Running disk benchmark tests against the RAID
: sets and against standard disks yielded very similar results.
 

: In other words, the saftey gained by RAID is worth the
: performance hit that you may see.
 

: - Al
: --
: Al Dorrington
: awdorrin_at_ictest.delcoelect.com Database Admin
: Delco Electronics - IC CIM Unix Sysadmin
: Kokomo, Indiana, USA Phone: 317.451.9655
Received on Tue Sep 10 1996 - 00:00:00 CEST