Semaphone problem?

[Mott Given]

I have a server which is waiting on semaphores about 17%-20% of its time, according to the Glance "B" report. The output of "sar -m" is
00:00:01 msg/s sema/s
01:00:01 0.81 173.86
02:00:01 0.81 150.87
03:00:01 0.81 1955.12
04:00:01 0.81 939.85
05:00:01 0.81 152.95
06:00:00 0.80 170.88
07:00:01 0.81 175.47
07:20:01 0.81 293.04
07:40:01 0.81 641.27
08:00:00 0.81 292.54
08:20:00 0.81 181.17
08:40:00 0.86 193.74
09:00:00 0.81 211.28
09:20:00 0.81 174.44
09:40:00 0.79 214.23
10:00:00 0.80 200.43
10:20:00 0.87 214.30

Is this amount of wait time for semaphores a problem? The system has a load average of 0.6 and was last rebooted on May 2. It has 25 ORacle databases defined in /etc/oratab. The system is an HP V2500 with 14 CPUs and 16 GB memory running HP-UX 11.00. The kernel values are:
sema 0
semmap 421986304
shmem 0
shmmni 1024

Mott Given

[Alex Lavrov.]

The columns are:
time of check
number of msgrcv() per second
number of semop() per second

How many semaphors do you have?
ipcs -s | wc -l

[Mott Given]

$ ipcs -s | wc -l returns 85.

Mott

[Alex Lavrov.]

What are the values of these parameters:
semmni
semmns

(The values you posted are for a shared memory segments, not semaphores)

[Mott Given]

The values are:
semmni 6440
semmns 12880

Mott

[Steven E. Protter]

You certainly are heavily using your semaphores, much more than I was with multiple oracle and other database instances and oracles application server.

I'd take a look at glance/gpm and see the percentage of semaphore use. If its high, I'd increase the afrementioned parameters.

SEP

[Alex Lavrov.]

On my busiest server, I see similar times.

The question is, is there a performance problem?

I know I have one, but the semaphores are the symptom, not the problem.

Are you short in memory?

Alex.

[Mark Nieuwboer]

Hi,

The default settings for oracle is that the database can use 3000 processes. normaly you don't need that you can reduse this to 300 or less after this you don't have any problems with your semaphores.

grtz. Mark

[Mott Given]

THe system doesn't seem to have a memory problem - the paging rate is generally under 20 pages/sec.

Mott

[Mark Nieuwboer]

Hi Mott,

I had the same problem an a oracle cluster.
i couldn't get all my databases up because i run out of semaphores. The oracle dba'ers of my company have lowerd down the number of process in the file init_std.ora.

after the restart of the databases we had no problem with the semaphores.

grtz. Mark

[Alex Lavrov.]

The problem is not number of semaphores, the problem is that the creation/access rate to semaphores is too high.

My opinion is that there is little you can do about it. Your DBA is the key to the solution. Suggest him to go over the oracle parameters. I know that DBA's don't like when system administrators tell them what to do ;-)

Alex.

[Mott Given]

Our ORacle DBAs are looking at this, but they do not think that there is a semaphore problem.

Mott