Semphore Settings on Sun V10

[Tim Wanner]

On bringing up some 30 sites and hundreds of processes on a SunFire V1280 12X900Mhz 24 GB, we ran into errors with shared memory, and not enough semaphores, with the folowing error messages, that caused the monitor daemon to panic.  We understood the semaphore setting on V10 Solaris would be configured dynamically, as the resources were needed.  Apparently not the case. Would anyone like to share the settings they are using for V10 Solaris?

oilMsemInit: semget failed, errno = 28

Semaphore file : /opt/quovadx/qdx5.5/integrator/testnrz/exec/sem_400022ac       Key Generated : 620793786

[msi :msi :ERR /0:  hcimonitord:07/30/2007 10:33:34] msiTocInit: Can’t init thread nrz4orbit_in semaphore: No space left

on device

[prod:prod:ERR /0:  hcimonitord:07/30/2007 10:33:34] Initialization of global shared memory failedPANIC: “0”

PANIC: Calling “pti” for thread hcimonitord

Topic

[Bill Bertera]

What version of Cloverleaf do you run? We have this problem, and I had it described to me as a bug in Cloverleaf, which I’ve heard is now fixed. Its due to the number of sites you run.

Basically, as it was described to me, is that there is an algorthm in Cloverleaf that assigns a semaphore key based on the name of the site. Apparently it isn’t too smart, and can assign the same semaphore key to multiple sites, so when the second site starts up it is not unique enough and you get that error. It is not actually a system-semaphore issue.

[Tim Wanner]

Unix Sun server with v10 Solaris, and v 5.5 Clovleaf.  Does anyone know the correct settings for Solaris 10?  These are the default settings, but do not work with Cloverleaf.

semmni  = 8192

semmsl  = default (512)

semopm = default (512)

semmns = obsolete

[Author]

Replies