Re: Oracle CPU/Mem/Disk config recommendations

Some addtl' info:

The new servers are for a different group of Iman/Unigraphics users who are currently using a 1000 for oracle and file storage. I am using the other 4500/450/250 setup as a baseline for comparison; I inherited both of these configs and when upgrading the group using the 1000, I don't want to make the same mistake the admin who upgraded the other group made.

 During mass imports to migrate a group of standalone UG users some months ago, our IMAN support team complained about poor performance from the 4500 oracle server. We ran MASSES of performance stats, vitual Adrian, all that jazz and armed with an inch thick binder of data, finally came to the conclusion that we needed to tweak the shared memory settings in /etc/system (which we did) and to get swap off of the slow diskpacks as well as get rid of the raid5 on the database disks (which we have not done yet because it was decided not practicle at this time).

While importing the data, most of the Oracle transactions consist of "does this file exist" with a reply of "nope". I'm not an Oracle DBA in any way, shape or form, so I don't know how to tweak whatever scripts or queries are run. Iman is pretty much installed by the folks who support the app, and they don't do any tweaking of the database or application in any way.

Which leaves me in bind. I don't have any control over the database or application, so all I can do is make sure the hardware/OS perform as good as possible with this information.

Thanks again for any suggestions or pointers to Oracle/Sun docs on this issue. Can't seem to find much on Sunsolve, and all the good stuff on Oracle appears to require an account to access!

-krp
"Kevin R. Pacek" <kpacek_at_excite.com> wrote in message news:gP9E6.12127$gl.1628732_at_typhoon.nyroc.rr.com...
> I will soon be tasked with building 3 new servers for Unigraphics. I
> currenlty have a 250 for imanweb, a 450 for the file server and a 4500 for
> the Oracle database.
>
> The new boxes will all run Sol8 (our IMAN team says this is ok), and have
> decided on what to use for imanweb and file server, but have some
 questions
> on the oracle server. When doing large imports on our current 4500
> (Sol7-32bit, 4x400mhz, 4gb RAM, 72gb RAID5 on A1000, boot disks and swap
 on
> OLD slow 7200rpm diskpacks) there have been complaints about performance,
> although day to day use by about 100 users is fine. We made all requested
> changes for Oracle and for shared memory tweaks in /etc/system. Iman
> support says it's configured optimally.
>
> My current spec is for a E4500, 4x450mhz, 4gb RAM. I was planning on
 using
> a D1000 with 4 36gb disks (10000rpm), two for mirrored root, 2 for
 mirrored
> database. Our future database needs will never exceed 20gb. I'm also
> planning on using the 64bit kernel.
>
> Questions 1: I may have the option of using a 3800 with 750mhz
> UltraSparcIII chips, but I would have to go with 2 to stay on budget.
 What
> will the performance of the 2 faster processors compare with 4 slower
 ones?
> I'll still have 4gb RAM.
>
> Question 2: I have the option of using the D1000 only for boot disks and
> using an A1000 for the database disks. This would allow me to use the
> hardware mirroring available with the A1000. Would the performance
 increase
> justify the considerable increase in cost? I'm guessing no, but would
 like
> your advise. BTW, I'm aware of the redundancy benifits of the A1000's
 dual
> controller, but I'm quite confident in the reliability of the D1000/ diff
> scsi combo.
>
> Question 3: Should the disks be newfs'd any special way for optimal
 Oracle
> performance?
>
> Question 4: I've read that swap should be distributed across spindles,
 but
> shouldn't I add more RAM before worring about swap config? If not, should
 I
> buy more disks and stripe/mirror (Raid 1+0) them for the databases and
 swap?
>
> Thanks for any input and literature you can point me to. I've read thru
> Unigraphics Solutions' IMAN sizing guide for Solaris, but It doesn't get
> specific enough.
>
> Thanks
> -krp
>
>
Received on Sat Apr 21 2001 - 08:18:36 CDT