Tunning question

[Vogra]

Hi all,
I know it's dificult to get a solution to performance problem with don't know all about the system, but I have a question...

My system (rp2470 - 4G RAM - DS2300 3 disks mirrored) is running one Oracle (8.1.6)instance with 340M of SGA (oltp and batch) with 250 users (concurrent).

Well, some parameters monitored said the following:
sar -c 3 20

***** SYSTEM CALLS *****

HP-UX vsurjux5 B.11.11 U 9000/800 07/21/04

19:02:54 scall/s sread/s swrit/s fork/s exec/s rchar/s wchar/s
Average 7354 1680 235 0.22 0.18 49941051 13773

the question is, it's says something about system calls? I read that it should have to stay over 1000.


Thanx in advance.
Lima.

[Sundar_7]

hmm..that is news for me. I have never heard of such thing that number of system calls per second should be more than 1000.

moreover, number of system calls/sec is the last thing that comes in to my mind while troubleshooting performance issues.

Start with the basics - are u sniffing a disk or memory related ?

Start with top, swapinfo, vmstat, sar and take it from there.

[Hein van den Heuvel]

>>> I know it's dificult to get a solution to performance problem with don't know all about the system, but I have a question

Good to see you acknowledge that.

Like Sundar said... what problem do we _think_ we have. CPU? IO? Memory? All-of-the-above? Forget system calls/sec.

From what you wrote I'd like to make the conclusion that you are sort of implanting in the question:
- The SGA buffer area is on the low side.
With 250 MB buffer (assuming at least 100MB shared pool) for 250 sessions: that's only 1MB/session. 250 MB buffers for 4GB memory... that's less than 10%.
Bump the buffer pool size to 500Mb (still lowish) and compare before and after stats!


- The app is doing tablescan's

Here is why: 50MB/sec with 1700 IO/sec. That's about 30kb/IO... more than your average single page read. Carefull check UTLxSTAT or STATSPACK output to try and nail down the expensive queries and tune those (maybe lock a table or two in cache)

Good luck,

Hein.

[Con O'Kelly]

Hi

As the others have already said you need to try and elaborate a bit more on what you think is the performance problem. Is it CPU, Memory, Disk?? Are Oracle users complaining about poor response time??
At this point forget about looking at system calls, check the basics first as follows and post back if you wish:

CPU: sar 2 10
Queues: sar -q 2 10
Memory: vmstat 2 10
Swap: swapinfo -tam
Load: uptime

As Hein has already pointed out, your Oracle SGA appears on the low side, especially considering you have 4GB of Memory. I'd also check that you have configured a reasonable system buffer cache - around 400M-500M.

In many cases the performance issues are related to the Oracle/Application configuration rather than the system itself.

Cheers
Con

[Vogra]

Hi Sundar,
I said over, but correct is under. Escusas.

Well guys, it's machine is running for 109 days since it's was started.

uptime
9:37pm up 109 days, 12:11, 2 users, load average: 0.47, 0.50, 0.52

nor swap and i/o. I think (monitoring) that is problem with code of new system.
Only CPU is high. I think the problem is poor code. But I don't know about aplic tunning... I continue monitoring...
If I have news, I will post here...
thanx...

[Dave Chamberlin]

Hello,

What applications are your users running that hit the database? How large is the database? Can you give more details about your disk system, in particular the disks that store the Oracle data? You say you have 3 disks mirrored - how are they mirrored? How many CPUs are in the system? How long has the system been in use - ie have you had the system for years and its now having problems, or is it a recent setup that has had problems from the start?