Cannot change SEMMNI kernel param...

[Coolmar]

Hi,

I need to up the value of SEMMNI from 1024 to 4096. Every time I do, the kernel rebuilds, then reboot and CRASH. I have to recover and move the old kernel into place. I have tried building manually and through SAM, same result with both. Any ideas?

Thanks!

[Pete Randall · Accepted solution]

The maximum value for SEMMNI is limited by the amount of memory available to you. I suspect that 4096 is too high. Try doubling it to 2048 and see if that works. If it does, keep increasing it until you find out what your max is.


Pete

[Steven E. Protter]

Shalom Sally,

You may have a formula elsewhere in your kernel causing trouble. Pete's suggestion is a good one.

Bill Hassell suggests removing all formulas from hpux kernels.

Hi Pete.

SEP

[Coolmar]

There should be lots of memory available. So it seems that I might have to find all the kernel params that are affected by SEMMNI and alter them, if they have a formula?

[Pete Randall]

Some of the relationships:

semmni = semmns =<(semmni xsemmsl)
semmap =<(semmni + 2)

There may be others, but, if you're using SAM to build your kernel, it shouldn't let you violate any of these.


Pete

[A. Clay Stephenson]

You are probably violating some of the other values
e.g. semmni < semuns

If you will will display the values of all the semxxx tunables the problem should be obvious.

[Pete Randall]

Oh, and

Hi, Steve!


Pete

[Coolmar]

sema 1 - 1
semaem 16384 - 16384
semmap 1026 - (SEMMNI+2)
semmni 1024 - 1024
semmns 32767 - 32767
semmnu 4092 - (NPROC-4)
semmsl 2048 Y 2048
semume 512 - 512
semvmx 32767 - 32767

[Steve Lewis]

The kernel allocates memory according to the sem___ values in the system file. I have run with 4096 semmni before, but not with such large undo structures as you. I suspect the undo structures may have something to do with your problem, given the number of semaphore ids you want.

In your case you have specified ((24+(8*512))*4092) (=about 16Mb) for the undo table alone.

Look at docs.hp.com for document
Commonly misconfigured HP-UX resources.pdf which will tell you good guidelines for configuring your semaphores.

If you are on 11.23, or 11.00 with PHKL_26136, or on 11.11 with PHKL_26183 your semaphore table will be dynamic, so you will be better off. If you are still on <=10.20, then you've got no chance with those values.

Read the above document for more info.

[Coolmar]

I need the semmni at 4096 for Oracle. Once Oracle is installed I will bring that and some of the other parameters back down as Oracle requires the params to be much higher than they really need to be. I am running HPUX 11.11.

[Coolmar]

Oh, also what are "large undo structures"?

[Coolmar]

I don't think it is the sem* values. I just tried to install the above recommended patch to make the kernel dynamic for the sem* portion and it crashed. Seems I can't change the kernel at all or it crashes. Is there a way to find out why or what is causing the crash?

[Don Morris_1]

Posting the panic string would be a start.
(Cut and paste it from the console log, or if there was a dump - go to /var/adm/crash/crash. where N is the largest number (and your most recent crash) and look at the INDEX file there -- it should have the panic string included and you could attach that to a message here).

Better yet would be contacting your HP support rep and getting them to look at the dump if there is one or see if this is a known panic case.

[Don Morris_1]

Oh, and I'd make *very* sure that the patch build wasn't trying to pick up your old tunable values. Given the way tunables are rebuilt into the kernel in 11.11, I wouldn't be at all surprised if your master and system.d files are including the semmni value you were trying to get to in the first place which may be causing problems in the patch rebuild.

If it were me, I'd reset any tunables you were trying to modify (using SAM or whatever you used before) back to "safe" values and see if that kernel works. Then patch that kernel... and then go to the increased values.

[A. Clay Stephenson]

I kinda, sorta suspected this from the start. I assume that you are booting from vmunix.prev or another old kernel and you are taking care not to overwrite this good one during kernel builds. This often happens when a kernel or network patches fails to completely install or all the co-requisite patches were not applied. Do a swlist to check the last swinstalls and then examine /var/adm/sw/swinstall.log.

[Coolmar]

I am going to log a call with HP. The system is HPUX 11.11 and Oracle 10g is running on it. It has 4GB of memory and 4 CPUs. There is only one database on it and the SGA is 600mb. Can any of you give me some guidelines/values as to what these kernel params should be set to? I would really appreciate seeing what others have.

thanks!

[Peter Godron]

Sally,
attached my list of kern params.
HP11.11 , L1000 , running 1 small Oracle10 DB, 512Mb RAM