HP UX @ORACLE 10G

[ALIREZA_5]

we have an HP RP5470 with 4 X 750MHz CPUs , 6GB RAM, 4 X 36GB LVD HDD. We installed HPUX 11 i on this server and then installed oracle 10g on this machine. However, with only 400 users connected to the server , the CPU Ideal gets to 0%, that I beleiv is not acceptable on this HP server performance!! Any idea where we might be going wrong?

[Alzhy]

400 Users for a 4x750Mhz, 6 GB RAM (what about storage?) system? I would say not Bad.

If you are seeing 0% idle time AND your CPU spends its time doing system and user (and not wait on IO's -- do a sar -5 100) AND all your processes are VALID.. then your box is not the right box for your load. Get more CPUs or go to a higher end server.

[Fred Ruffet]

?!

And why wouldn't it be acceptable ? CPU 0% Idle doesn't mean your machine is full. It just mean that the 4 CPUs are being used. Note that CPU usage report is not, like on windows, limited to 100%. Load can be very much higher.

Are user complaining about performances ? Is their a real bottleneck on CPUs ? Have you tried to use glnce (gpm) to monitor your system ?

Regards,

Fred

[Jean-Luc Oudart]

Hi,

do you the threshold (number of users) when the system is cpu bound ? what about memory ?
Could you post the kernel parameters and init.ora file ?

Regards
Jean-Luc

[Patrick Wallek]

What does your memory utilization look like? Hopefully you have enough RAM and are not paging.

# swapinfo -tam

Oracle 10G and 400 users, 6GB doesn't sound like much. Post the results of the above swapinfo command and maybe your kernel parameters so we can look at those.

[Steven E. Protter]

There is absolutely nothing wrong with running a Unix system at 100%.

The real question needs to be, is there a performance problem? You may wish to collect some performance data. I'm attaching a set of scripts.

Are you swapping?
swapinfo -tam

100% is simply not a problem that needs to be dealt with.

Other specific symptoms should be dealt with.

I'm implementing 10G here right now. I'm interested in your memory use profile right now.

See script attachment.

SEP

[Alzhy]

100% CPU Itilization (specially if its more USER thn a SYSTEM activity) will mean you could have trouble ahead. It may not currently (or most of the time) be impacting (yet) but it looks to me that based on the system's configuration and "only" 400 users that this configuration will be enough.

This added to the fact if memory, IO and other environmentals are fine.

[ALIREZA_5]

Steven, Thank you for your reply, attached is the result of the script you requested us to run.

[Alzhy]

Alireza,

Can you provide "sar -d 5 100" and "vmstat 5 100" output when your system is at "0% Idle"? It may be that your 0% idle does not really mean the CPU is 100% busy -- it may mean you have a lot of wait IO's - which judging by the data you gave us .. you are only using the 4 SCSI disks that are on the coreIO SCSI of your 5470.

Correct me if I am wrong but you seem to utilize 1 36G disk for your OS and another 36 G for your /db (is this your oracle software tree?). And it is possible you use the other 2 36 GB as raw oracle storage or you did not use them?


Man 400 users and Oracle 10G on slowww SCSI storage? That does not sound right? You have a FC HBA.. why not connect it to fast disks?


OIther things. You're way behind in your HWE and GOLD support bundles. The latest is June 2004 (I think).. Also, hows your kernel settings ? Can you also send "cat /stand/system" and "ipcs -a"?

[ALIREZA_5]

Steven, Thank you for your reply, attached is the result of the script you requested us to run.

[Steven E. Protter]

I'm looking at your perf docs, though I'm having trouble right now. I'll get back to you on that.

Since you use 10G can you take a few minutes and read this thread for me:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=734418

I'd appreciate your input since you are obviously at a more advanced state with this product than I am. If you post a useful response to my thread, of course you'll be rewarded with points.

Thanks.

SEP

[ALIREZA_5]

DEAR ALL

I RUN THE SCRIPT AND THE FOLLOWING RESULT
WAS OBTAINED

PLEASE LET ME KNOW WHERE I SHOULD BE LOOKING
TO RESOLVE THIS PROBLEM

[Alzhy]

Alireza,

The stats you provided seem to have been taken when your system was not busy - what happened to the 400 users? On one stat.. it did seem to confirm my suspicion that at times that you've seen "CPU Ideal gets to 0%" .. you have very significant WaitIOs.

Read my previous reply above... And confirm where do you store your DB.. you have 4 "slow" 36 G disks hanging off your internal SCSI bus. 1 is used for OS the other is used for /db -- what are you using the other 2 36GB disks for? Are you using them as Oracle storage? This environment alone - IS NOT A GOOD DATABASE storage environment.

Post the other stuff I've requested.

[Steven E. Protter]

You may need to run the scripts a little longer.

The CPU information you posted shows idle time.

perhaps collect data for an hour.

I name the script /usr/contrib/bin/system.perf.sh

run it as
/usr/contrib/bin/system.perf.sh 3600

runs it for an hour

SEP

[ALIREZA_5]

Dear all,

The 400 users connect to a web logig App server and the app server connects to the DB server.One other thing that we have done , we have moved the DB to a Virtual Array 7110 storage (5 X 72GB FC 10K HDD).

[Alzhy]

Okay..

So WHEN did you do the move? Is everything fine now? If not please provide the other info I am requesting...

[ALIREZA_5]

Dear all

do you all agree with the patches that i have applied.

i have a feeling that this problem is due to
some kind of software tweaking

[Steven E. Protter]

Further analysis:

You seem to be waiting on disk i/o, Not as bad as my systems, but you have a better disk array than I.

Possible causes:
inefficient sql in the ias application. That can create lots of database io and significant system drag.

Problems with disk layout of the database and/or application.

The application is pretty much execute once, but 10g does have a psuedo database associated with it and that database can have issues.

Are you aware that Oracle recommends Raid 1 or Raid 10 storage for databases and redo logs? If you don't do that you can have performance issues, even with few users.

SEP

[Alzhy]

Alireza,

I for one am not content with your level of Patching. JUNE 2003? That's too old..

Also, if you cannot accurately give us your environment - we cannot be effective in helping you.

[ALIREZA_5]

Dear all
Just to let u all know, that the App servers are HP-Compaq ML530-G2 with 2X Xeon 2.8GHz, 4GB RAM. Would this cause a performance issue?

[ALIREZA_5]

ANY THOUGHTS?

[Jean-Luc Oudart]

Hi,

as per my previous post could you post kernel parameters and init.ora file.
Also, would you have a statspack report ?

Regards
Jean-Luc

[Steven E. Protter]

as root

kmtune

displays all kernel parameters.

Did you re-run the performance scripts again.

What I have seen so far does not show the problem you originally posted on.

Because you're new, take a look at this document at some point.

Off thread topic
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=140137

SEP