freebsd stable - Mar 2004 - Semaphore scaling

I am running an application (Plesk) on a 4.9 system that is eating up all the
semaphores and as a result some processes can?t run. I have contacted the
software company, and they asked me to increase the semaphore limits in the
kernel. However their suggested settings seemed very unproportional to me.

So my question is, how should I adjust the values, while keeping things in
proportion? I wish to increase the default semaphore limit of 10 to 20. Can I
simply double the semaphore related values?

This is what the support guys suggested:

---

# number of semaphore identifiers in the system - default is 10
options     SEMMNI=256

# other things that have to be set to allow increased semaphores without
destabilizing system

options     SEMMAP=768      # amount of entries in semaphore map
options     SEMMNU=768      # number of undo structures in the system
options     SEMOPM=2560     # max number of operations per semop call
options     SEMUME=256      # max number of undo entries per process

# number of semaphores in the system - default is 60
options     SEMMNS=393216

# max number of semaphores per id - default is 60
options     SEMMSL=1536

---

Thanks.

1

Don't crosspost lists. Thanks.

On Thu, 4 Mar 2004 mikej@rogers.com wrote:
> I am running an application (Plesk) on a 4.9 system that is eating up
> all the semaphores and as a result some processes can?t run. I have
> contacted the software company, and they asked me to increase the
> semaphore limits in the kernel. However their suggested settings seemed
> very unproportional to me.
>
> So my question is, how should I adjust the values, while keeping things
> in proportion? I wish to increase the default semaphore limit of 10 to
> 20. Can I simply double the semaphore related values?
The semaphore tunables aren't quite as obvious as they look. There are
limits on the size of various clusters of things whos usage is heavily
application-dependent. That being said, FreeBSD traditionally has very low
SYSVSHM/SEM/MSG limits, so doubling them should be safe.  I suggest using
the ipcs(8) command to monitor the usage and adjust accordingly.

The Plesk-supplied limits seem OK to me too.  I've put Oracle-sized
values in there and haven't had any bad effects .. yet :)

Note that you can adjust these with loader tunables as well, if you don't
want to rebuild the kernel.  They have the same name as the sysctls that
show the current values in kern.ipc.

-- 
Doug White                    |  FreeBSD: The Power to Serve
dwhite@gumbysoft.com          |  www.FreeBSD.org