Kernel Parameters.

[Sundararajan Anandhan]

Hello...,

I have a HP K-class box running HPUX 10.20.
We have 'tuxedo' application running on our system. Recently the number of tuxedo environments have increased and I am hitting 80% of 'msgmni'.

It is currently set to 650 and I am thinking of increasing it.How is it going to affect the performance of the system..? Can anyone help me on the effects of increasing the parameter...? Also on increasing 'nproc'.

Thanks in Advance.
Sundar.

[Stefan Farrelly]

There will be no noticeable performance increase by increasing the number of message queue identifiers, or increasing nproc. The kernel will use a little more memory for its kernel tables, thats all. Increasing nproc will allow more processes to run (if youre not using this to keep the number of processes down) which may put a more noticeable load on your server.
ITO is the same, you need to increase msgmni well above the default as it uses them for communication.

[CHRIS_ANORUO]

Hi Sundar,

Have a look at doc, it help to understand the parameters to change and others are dependent on them:
http://www.docs.hp.com/hpux/content/KCparams.OverviewAll.html


Chris

[Manju Kampli]

Sundar,

Allocating Memory Space for Messages
IPC messages require the following memory space allocations:

Space in the kernel area for message identifiers.

Shared memory space for message queues.

Shared memory space for message headers.

so plese look in to your memory availability of the system.

[Anthony deRito · Accepted solution]

Tuxedo uses the IPC facilities. If you are going to increase the number of running Tuxedo processes, there has to be at least one message queue available for each process.
However, it would be wise to check the currently allocated message queues on the system and see if there are any "hanging" around on the system.

This problem is typical for Tuxedo when things go wrong. Especially if your not running 6.x. Use the command:

#ipcs -q | grep [tux_app_id]

This is an indication of the number of currently allocated message queues. You may want to first clean up these resources with the ipcrm command, restart your processes then check out the table again.

A clean reboot will also clear out any message queues not in use.

Unless your very good with tmadmin, things will happen that get you in this situation.

Then, when your confident that this is OK, go ahead and increase your msgmni parameter.

Note that the nproc table is not directly related to this IPC issue. Of course, at a system wide level, nproc needs to be large enough.

Tony

[Sundararajan Anandhan]

Thank you all for responding..

Tony , I did remove the orphan queues . Since the developers are creating their own little test domains, the usage has increased. I have increased it to 1024 and its fine now. Thanks for your inputs.
I have a question for you Tony, What is the 'tmadmin ' that you are talking about. I didn't quite get it.

Thanks,
Sundar