Hi!
We are about to plan our next database server.
(Linux RH 3, Oracle 10.2.0.1.0)
It's got 6 discs, so it's a pretty small system
and we have tentatively decided to go with oracles
S.A.M.E. approach, namely as much mirroring as
necessary and as much striping as possible. And
put the slowest data (probably OS and archived logs)
on the inside and the performance critical data
on the outside of the discs.

So, we plan two mirrored groups. OS inside, then
archived logs, then - yes, that's my question.
How is I/O load normally for datafiles control files
and redo logs during an insert?

Our typical load scenario is a check program using
oracle spatial.

Lots of data gets inserted into different tables using
sql*loader conventional path (typically at least one
512MB redo gets full), statistics get collected, for
our five complex queries get executed, which churn
through the data just inserted (mostly spatial data),
return very little data (check results), and then
the data gets deleted.

This happens about 40 times a day (four users, 10 times per
user).
For concurrency reasons I can't mess with the indexes, they
have to stay active all the time.

We've got 2G of RAM, so I can configure a large buffer cache.

I'm not really sure where there's more load, on the redos or
on the data files. I guess, control files aren't critical as
they only get written after each commit, don't they?

Lots of Greetings and Thanks!
Volker

On Mon, 3 Apr 2006 16:05:17 +0200, Volker Hetzer wrote
(in article <e0ra2t$tc0$1 (AT) nntp (DOT) fujitsu-siemens.com>):

Volker,
for the storage I would go for ASM. Have you considered this ? It really is
worth taking a good look.

--
With kind regards / met vriendelijke groeten,
Ronald

http://ciber.nl
http://homepage.mac.com/ik_zelf/oracle

Ronald Rood wrote:
database instance and I have to keep OS, SW and the ASM separate from
the database. This leaves me with four disks instead of six, therefore
losing a third of the write performance.

Is there a way on linux to monitor which file gets written now often
and with how much data?

Lots of Greetings and Thanks!
Volker

Ronald Rood schrieb:
You were right.
After I figured that ASM can work with partitions we found a way.
My current idea is:
- We combine our six discs into three drives (Mirroring), using
RAID 1 on the controller
- partition each drive into
* Non-DB (Linux, Software, et.) (innermost cylinders)
* Db slow (archived logs)
* DB medium, fast
* DB Really fast (control, redo) (outermost cylinders)
- and let ASM do the striping across the three drives for
the DB stuff. That way we don't have to buy an expensive
RAID 10 or 0+1 capable controller and my boss is happy.

Lots of thanks for pointing me to ASM!
Volker