Advices needed: what common value for maxusers

[nicnictout]

Hello,

i have a OVO 8 on HP-UX, and yesterday, manager crashed because my kthread table is full.

I saw that i need to up the maxusers, nprocs and nfile values. My question is, this issue occurs twice, so i would like to know,

- what is the max value can i put for theses parameters

- what impact on my system, high values will make

I have two OVO in service guard cluster, supervising more than 1000 servers and network devices (through NNM).

Thx for for help

[Gino Castoldi_2]

Hi,

- what is the max value can i put for theses parameters

These values depend on different factors such as the system's memory, etc. so there is no one answer to this question.

- what impact on my system, high values will make

You could make your servers non-bootable and/or cause other system issues if they are set incorrectly.

I would recommend posting this question to the HP-UX Forums (System Administration) as they have the expertise to help answer your system questions.

Here are two links that may be helpful:

http://zensonic.dk/?cat=9

http://h30499.www3.hp.com/t5/System-Administration/vmunix-kthread-table-is-full/td-p/3755367

HTH

-Gino

[nicnictout]

Ok thx very much for your advices :)

[Bill Hassell]

maxusers is absolutely useless and was created at a time when RAM was in megabytes, not gigabytes. Back then, it was assumed that each user consumed a certain amount of RAM, certain number of processes and a certain amount of shared memory segments. It was a poor guess at the time and is totally meaningless for servers today. For example, a database server or web server has virtually no users -- all the activity is through remote communication.

So formulas listed in sam or smh (or kmtune/kctune) are starting points that must be adjusted to match the applications and usage. kthreads must be made extremely large for Java applications like NMM OVO. If you are running out of threads, replace the brain-dead formula with a real number. If it is currently 1000, make it 10000. Yes, 10 times larger. This has no effect on the running kernel except to make some space to track each thread. Since you no control over the actual code (NMM, OVO), you need enough thread slots for your busiest condition, now and in the future. For nfile and nproc, use sar -v 1 1 to see how close you are to the limits (for HP-UX 11.11) or kcusage for HP-UX 11.23 and later. To see the last month's usage, use kcusage -m nproc nfile

Increasing these parameters will require more memory for the kernel but not increasing them will cause the applications to fail. For a system this extensive, be sure you get a copy of GlancePlus so you can actually monitor your system and parameter usage.