High shmctl system calls - what is the source process

[Bernie Vande Griend]

We are having a performance issue with our Filenet IS app running on HP-UX 11.11 which we've determined has a very high # of system calls. From glance I can tell these are all shmctl calls but I'm having trouble identifying which processes they are coming from. I ran a process Measureware report but didn't see it in there either. Any ideas are very welcome.

[Steven E. Protter]

Shalom,

UNIX95=1
ps -efH | grep process_name

-H with UNIX95 set shows process hiearchry. The child processes are indented.

Also of possible use is lsof which provides information on processes.

Lastly you might wish to try tusc, which will show you everything a process does if you hang a tusc tracer on it.

sEP

[Bernie Vande Griend]

I don't see anything in lsof that will show me system calls for a proc and the ps option as well. I could use tusc, but there's hundreds of processes on here so I'm really looking for a tool that I can run against all processes on a server to see what there system call counts are during a time period. glance comes close, the Y option shows me the total system calls and the L option shows me counts by process. If I could do the -L for all procs I'd be all set.

[Dennis Handly]

>I can tell these are all shmctl calls

Do you know if they are SHM_* or IPC_* operations?

>I could use tusc, but there's hundreds of processes on here

You can use tusc with -fp to follow a whole process tree from the start.
You can also pass in 100s of PIDs to tusc.

[Bernie Vande Griend]

I ended up opening a support ticket on this and HP performance tech had me run a kernel trace with kitrace for 20 seconds which they then analyzed to help us figure out there the system calls were coming from. They were coming from over 600 different procs as children proc of the apps main proc. Still working with the app vendor on this. Now looking for a method to alert us when total system calls reach a certain limit.