parameter settings

[Martin Wells]

We are currently making some changes to our tunable kernel settings, However when we are trying to make the change SAM will not allow us because some other settings have already been change to values that it does not allow.

These are

max_thread_proc <= nkthread
and we have
max_thread_proc = 20000
nkthread = 2048

ncallout >=nproc
and we have
ncallout = 2064
nprocs = 16820

Does anyone know if it will be a problem to change these settiings to ones that SAM allows?

This would mean our know values would be

ncallout = 16836
nkthread = 29451

The system these settings are on is a L class running HPUX 11 with three processors and 16gb of memory.

Thanks

[Chris Wilshaw]

There shouldn't be a problem with increasing those parameters to allow you to increase the ones you want, especially with that hardware setup.

There are also similar restraints on the values of other parameters (the sem* set for example)

[V. V. Ravi Kumar_1]

hi,
some kernel parameters values depend on other values. they may have formulae, which involve other parameter values. so change values in an order by observing the messages.
regds

[T G Manikandan]

Just do the kernel modifications from the command line for the values which are improper with SAM.

ncallout should be nproc

so you can modify those values and do a build

# cd /stand/build
# /usr/lbin/sysadm/system_prep -s system
# vi system
Make your changes
Save the file
# mk_kernel -s system
Once successful you need to put the new kernel in place.

# mv /stand/vmunix /stand/vmunix.prev
# mv /stand/build/vmunix /stand/vmunix
# cd /
# shutdown -r -y 0



Thanks

[harry d brown jr]

You have a limit on the number of processes your OS can handle, like around 30K. I'm not too sure why one would need to allow ONE process to HAVE 20000 threads (max_thread_proc)?? You'll also have to increase the number of files allowed to be open (nfile) and inodes (ninode).

live free or die
harry

[keith persons]

noe note - ninode should "only" be increased if, and only if, you are using HFS file sytems extensively - if all your file systems other than /stand are vxfs hardcoding ninode to 3,000 or less is highly recommended.

keith

[Martin Wells]

Our ninode setting is

ninode = 19088

Most of our filesystems are Vxfs

Is this going to be a problem??

[keith persons]

Martin,

a value of ninode is not the concern it once was though I would recommend reducing it to 3,000 on the high side, 1,000 on the low side. JFS file systems have their own kernel identifier and it is not only dynamic but also a linked list - ninode has no affect on the JFS kernel inode structure.

Keith

[Martin Wells]

Keith

Would having this value set so high have performance issues

Thanks

[keith persons]

Martin,

It probably won't cause any discernable problem because the processor(s) is/are quite fast and improvement in some of the kernel table management have greatly reduced the impact of large values of ninode. If anything, you may have a little extra system mode on the processor util, but I don't think it would noticeably degrade throughput.

keith