Shared Memory Table fills up with Cognos

[Michael Geldart]

I have an RP8400 with 8GB of RAM.

I am running 3 Oracle 9i databases and Cognos Enterprise Series 7.

Every 2 weeks I get an alert that the Shared Memory Table is nearly full. Cognos ceases to function and I end up re-booting the server.

Has anyone seen this issue with Cognos?
Any suggestions on how I might free the Shared memory table without re-booting?

Thanks,

Mike

[A. Clay Stephenson]

I suspect that your fundamental problem is improper session termination; user training is probably called for.

You might try this:
Have a cronjob that does an ipca -ma and look for NATTCH = 0 and an ATIME (time of last shm attachment) older than some arbitrary value. You also need to filter by owner and/or group. You then call ipcrm using the shmid of any shmid's meeting the above criteria.

[Steven E. Protter]

kmtune | grep shmmax
kmtune | grep shmseg

to get current figures

kmtune -s shmmax=new_number
kmtune -s shmseg=new_number

I know that shmmax is dynamic under HP-UX 11.11(11i). If you are running 11.00 you need to boot.

All these instructions are in the following man page:

man kmtune

SEP

[A. Clay Stephenson]

Ooops,

that should have been "ipcs -ma" NOT "ipca -ma". Note also, the while NATTCH = 0 is a necessary condition for removal of a shared memory id, it is not necessarily a sufficient condition -- it depends upon how the application was written. That is why I said look at the times as well.

I rather doubt that adjusting the tunables is going to help; doubling the values will get you from two weeks to four weeks but the problem remains. I'm almost positive the real "fix" is either train your users to properly end their sessions (a baseball bat might be needed) and/or patches so that the applications do not crash. Finally, if you or other sysadmins use "kill -9" then "We have met the enemy and the enemy is us."