Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> RAID10 config...
Hi!
I'd like to collect some thoughts on raid10 io configuration
for a write-heavy system.
I've got 8 discs hanging off one controller which controls
2 SCSI buses. (9.2.0.4, RH linux) I can't have more disks
nor more controllers. The controller has 64MB of battery
powered cache.
I'd like to do mirroring on all files.
So, I can imagine several setups, the simplest being (IMHO)
1 pair for OS, SW (+ 1 controlfile)
1 pair for redos
1 pair for archivelogs (+ 1 controlfile)
1 pair for datafiles
My current one is
1 pair for OS, SW (+ 2 controlfiles)
1 pair for redos and archivelogs
2 pairs for datafiles
which I don't really like because all controlfiles
are on the same disk and ideally I'd have 3 controlfiles.
However, since the OS+SW won't create much load, I could in theory put the archivelogs there too:
1 pair OS,SW,archivelogs (+ 2 controlfile)
1 pair redos
2 pairs or 1 quadruple for datafiles
Am I right in assuming that during writes, controlfiles
become bottlenecks?
Would it more sense to spread out the datafiles or the
redos? The lamps on my redo disks are almost constantly
on during our load runs but I imagine that only small
portions are written each time. The statspack says 1,804.45
per transaction, 124,352.33 per second, I suppose it's bytes.
Also log file sync and log file parallel write together take up 55%
of the % total elapsed time in the top 5 timed events of my
statspack report. (db file parallel write are another 13%)
The problems I see are basically, how to distribute the controlfiles and how to make redo writes faster, given 8 disks.
Sorry, I forgot. Right now we've got each disk of a pair on another SCSI bus so that every access goes over two buses. I am thinking of reserving one bus for the redo disks and one bus for all the other disks. Would that make sense?
Lots of Greetings and Thanks!
Volker
Received on Tue Oct 21 2003 - 11:43:32 CDT