shared memory issues

[Gerard Moroney]

Hi,

I am running an application on an n-class cluster which uses an Oracle database. There is a particular process we run for which there is no problem when several are executed one after the other. When multiple numbers of these processes are executed simultaneously (different conditions, same process) some fail. We think it could have something to do with how shared memory is accessed/locked/released. Are there any standard hp-ux11.00 tools that could allow us to dynamically monitor shared memory and report memory errors ? I also have glance on the system.

[A. Clay Stephenson]

Hi,

Clearly the place to start is with ipcs -ma run at intervals. Probably of more value is to look at the oracle logs and look for errno values. If one of your processes dies, immediately issue an echo $? command to give the return value of that process. This is typically errno and can be of great value in tracking down the problem. You should also do an ipcs -a which will also give data pertaining to message queues and semaphores.
Your problem could simply be related to an inability to fork because of swapspace or nproc. It is probably a bit early to blame this on shared memory.

Regards, Clay

[Bill Hassell]

You may have a fragmented shared memory area, especially if Oracle is running in 32-bit mode. Remember that 32 bit Oracle is limited to a single 3.75Gb address space and that the same fragmentation which can occur in shared memory for 10.20 will occur in 11.0. When managing 32 bit apps, you might m=want to consider memory windows, or always start and stop the Oracle processes in the same order.