--
Frank
Doug Coan <dcoan_at_aegonusa.com> schreef in berichtnieuws
8qo5a3$3uc$1_at_nnrp1.deja.com...
<snip>
> - All of our best running DB that are in the 75-120 GB range are
> configured as a single raid 0+1 array and are spread out across 12+ or
> more drives. I know this goes aginst good DBA practices 101, but it
> really really works. I continously challenge different DB vendors and
> disk vendors to show me something faster and easier to administer and
> NONE have been able to yet.
Is that an open door or what? Usually improvements on one side
means more work on another.
Ever wondered why Oracle in the OLTP record breaking events uses raw?
That would perform even better - but is even less easy to administer.
Would said your life was to be easy?
<snipped some more>
> - Oh yeah - DON'T DO RAID 5 any where for any reason. I'm not talking
> nichel and dime performance gains either. I have NEVER gone from a raid
> 5 array to RAID 0+1 and not seen at least a 100% performance increase. I
> have gone back and forth several times to make sure it was just the RAID
> config and not any other type of fragmentation issue. I had one load job
> go from 39 hours to 3!!!!!
Ever did huge queries?
RAID-5 is - because of the way physical disks are combined to
logical disks- good in read thruput. Not write, as you have seen.
For some info on how to mix, what dangers may lure ahead, and what the
trade-offs are, I suggest you read
http://www.ipass.net/~davesisk/oont_performance_raid.htm
Alexander, hth, but be prepared to test several configurations until you
find
the optimal one for your load. It pays off to spend some time testing,
especially
with stripesizes!
Received on Wed Sep 27 2000 - 09:20:03 CDT
