kernel parameters on 11.0

[james gould]

What should the values for nproc , maxuprc
and maxusers be set to. At present they are
set to

maxuprc 75
maxusers 32
nproc 276

Are these set to low? Have a total number of
192 processes running.

[harry d brown jr]

It really depends upon what you are using the system for: number of applications, number of users, types of applications/processes, load, etc...

[linuxfan]

Hi James,

Well to answer your question "It Depends". If you want to get a general idea, look at /usr/sam/lib/kc/tuned. There are some files *.tun, which HP supplies by default for tuning your kernel parameters.

To get a better understanding about kernel parameters look at
http://docs.hp.com/hpux/onlinedocs/os/KCparams.OverviewAll.html

-Regards
Ramesh

[Sanjay_6]

Hi,

The others have already advised you a little. Like they said it all depends on what applications you are using. However i would advise you to increase tour nproc value to 1.5 times to be on the safer side. The 192 processes you referred to are max no of processes running or the usual no of processes at any given amount of time.

I would still advise you to increase nproc.

Thanks
Sanjay

[MANOJ SRIVASTAVA]

Hi James

I am attaching a sample o/p from one of the servers , actually this depends on the server users and database they are going to open.

Typicall you can start with mausers and then go around detrming the others.


Manoj Srivastava

[Wodisch]

Hello,

after telling you again, that it totally
depends on the work your machine has to
shoulder, I do recommend some kernel parameters
which are completely safe to increase (to the
values I tell you, at least):

create_fastlinks 1
max_thread_proc 128
maxfiles 160
maxswapchunks 1024
maxuprc 512
nclist 5120
nfile 3000
ninode 2000
nproc 1000
npty 600
nstrpty 600
nstrtel 600
swapmem_on 1

and set "dbc_max_pct" to a value of no more
than 300MB, nor less than about 50MB (I am
talking about HP-UX 10.20 or 11.00 - for 11i
you might go up to about 800MB)

That should be pretty safe, and keep your box
for some more time...

But anyway - tell us, what your box is doing
or going to do, so we can tell you something
fitting!

HTH,
Wodisch

[David Lieberman_1]

Hi James,

well... it depends...

Generally, I have found that the default kernel is unacceptable due to it's over-sizing of the Dynamic Buffer Cache (DBC) and under-sizing of process-oriented parameters.

I generally use JFS on all filesystems except /stand, a static buffer cache to avoid DBC contention with the vhand process, and adequate device swap to eliminate the need for pseudoswap, so my basic parameter list will differ somewhat from Wodisch's:

bufpages 76800 # 300MB static cache
create_fastlinks 0 # no HFS
dbc_max_pct 0 # see bufpages, no DBC
dbc_min_pct 0 # no DBC
fs_async 1
max_dsiz 4GB # 1GB on 10.20 (no larger than RAM. 1.75GB is possible on 10.20, but all future kernel builds with that value must be done by hand, or SAM will override and reduce it to 1GB).
maxfiles 200
maxfiles_lim 2048
maxssiz 79MB # (80*1024*1024)
maxswapchunks 8320 # 2xRAM + pri swap
swapmem_on 0 # Use RAM for programs, not to save disk space. a 9GB disk is a LOT cheaper than 4GB of RAM.
maxtsiz 1GB # (1024^3)
maxuprc 200
maxusers 124 # sometimes >128
netmemmax 0 # generally
nfile 2800 # often more
ninode <=5120 # <=4K on 10.20, <=15K on 11.x. smaller and functional is better
nproc 1024
npty 512 # sometimes 128, 256.

I try to modify params that affect formulas rather than editing formulas or *gak* overriding a formula with a plugged value.

As always, the above params are a starting point. They may not work for you, and they generally require some tweaking on our servers since each environment is somewhat different.

If you can, get a hold of the HP-UX Performance Cookbook by Stephen Ciullo, HP Senior Technical Consultant, and Doug Grumann, HP Performance Technology Center R&D Lead Engineer.

HTH
David Lieberman