Hitting "user limits" for IPC resources?

[Ty Roberts]

I am trying to investiage if a user is running into any IPC utilization limits, for all their process combined or for individual processes. I can use glance -t to see that globally, the system is only using 1 - 2 percent for shared memory, semaphores or message queues but could the user be hitting a process or user limit? I am at a loss on how to monitor or check for this. Can someone point me in the right direction? Also could they be getting IPC errors if they are hitting any of their ulimit settings? This users ulimit information is below.

ulimit -a
time(seconds) unlimited
file(blocks) unlimited
data(kbytes) 1048576
stack(kbytes) 131072
memory(kbytes) unlimited
coredump(blocks) 4194303

Thanks for all the help!!

Ty

[Bill Hassell]

You'll need to provide real error messages and the failing IPC resource. ulimit does not directly influence IPC categories. There's just not enough information yet.

[Dennis Handly]

As Bill says, you'll need to look at kctune(1m) 11.23 or kmtune to look at these kernel parms.
shmmax shmmni shmseg
semmni semmns semmnu semmsl semume
msgmap msgmax msgmnb msgmni msgseg msgssz msgtql
Each should have a man page on 11.23.

[Ty Roberts]

Sorry everyone, I didn't do the best job with my question. Guess that is what you get when you make an emergency post while troubleshooting a problem!!

We are running JD Edwards on HPUX 11.11, and it is getting some very generic "IPC related" errors like the one below.

1050 Mon Jan 7 13:42:41 2008 ipcpub.c2911
process table full for process 1050

The reason I say it may be IPC related is the "ipcpub" call that reported the error.

Even though the error is process table full, maxuprc and nproc limits are no where close to being hit. Along with that - shmmni, msgmni, semmni all have very low utilization (like I had mentioned). I was not sure if any of the ulimit setting had anything to do with IPC limits, so that cleared that up.. thanks!

I just don't have enough understanding of the other kernel parms to know a good plan on attack to see if the user or the user process is hitting any of those maxs. I have listed their values below to see if anyone has any input on adjustments.

msgmap 5122 - 5122
msgmax 32768 Y 32768
msgmnb 65536 Y 65536
msgmni 128 - 128
msgseg 20480 - 20480
msgssz 128 - 128
msgtql 5120 - 5120
sema 1 - 1
semaem 16384 - 16384
semmap 1026 - (SEMMNI+2)
semmni 1024 - 1024
semmns 2048 - 2048
semmnu 1024 - 1024
semmsl 2048 Y 2048
semume 512 - 512
semvmx 32767 - 32767
shmem 1 - 1
shmmax 0x40000000 Y 0X40000000
shmmni 1024 - 1024
shmseg 512 Y 512

Thanks for all the help guys.

[Bill Hassell]

> process table full for process 1050

Very useful. Although nproc seems to be OK, you probably measured it after the error message and the usage went away. You may need to monitor sar -v 1 continuously while the code starts until it fails.

Another possibility is that it uses a lot of threads (JavaJunk?) and you'll need to look at thread requirements for the JDE programs. Then compare with kernel parameters nkthread and max_thread_proc.

[Ty Roberts]

I don't think nproc was spiking, the limits are in the thousands and at any time that we were watching (while the errors were going on) the values were between 100 and 200. Now I guess the app could have been spiking tens of thousands of processes in a second, but highly unlikely.

JDE is not java, but I will look at the thread information. Thanks!

[Tim Nelson]

how about maxuprc ? nproc is global for the system, maxuproc is per process.

[Dennis Handly]

>I just don't have enough understanding of the other kernel parms

Try the man pages: (11.31 below)
http://docs.hp.com/en/B2355-60130/msgmap.5.html

[Charles H Emery]

Actually, if you google for the Kernel parameter name, usually the 1st or 2nd hit is an HP Doc for that parameter. The doc includes an description/explanation which is pretty understandable.

Example: I googled msgmap and got:
http://www.docs.hp.com/en/B3921-90010/msgmap.5.html

Cheers!

[Charles H Emery]

Actually, if you google for the Kernel parameter name, usually the 1st or 2nd hit is an HP Reference Doc for that parameter. The reference doc usually includes an explanation which is pretty understandable.

Example: I googled msgmap and got:
http://www.docs.hp.com/en/B3921-90010/msgmap.5.html

Works surprisingly better than using the ITRC search.

Cheers!