Kernel Parameter

[moodi]

Our customer asked us to change the kernel parameters.Here is the detail.

Server : RP7420
OS : HPUX 11.11B
Database : Oracle 10G R1.

The recommended kernel parameter are as follow,
Parameter Existing Recommended
aio_listio_max 256 512
dbc_max_pct 25 50
dbc_min_pct 5 2
ksi_alloc_max nproc * 8 nproc * 8
max_async_ports 50 1210
max_thread_proc 3000 7000
maxdsiz 2063835136 4294967296
maxdsiz_64bit 21474836480 21474836480
maxfiles 2048 2048
maxfiles_lim 2048 8192
maxssiz 134217728 134217728
maxssiz_64bit 1073741824 268435456
maxswapchunks 16384 16384
maxuprc 3687 ((nproc*9)/10)
maxusers 512 1200
msgmap 4099 2+msgmni
msgmni 4097 8192
msgseg 32767 already at max limit
msgtql 4097 8192
ncsize 35848 (ninode+vx_ncsize)
nfile 63503 (15*nproc+2048)
nflocks 4096 8192
ninode 34824 (8*nproc+2048)
nkthread 7185 (((nproc*7)/4)+16)
nproc 4097 8000
semmap 4098 (semmni+2)
semmni 4096 8192
semmns 8192 (semmni*2)
semmnu 4093 (nproc-4)
semvmx 32767 40000
shmmax 4294967296 12884901888
shmmni 512 1024
shmseg 120 512
swapmem_on 0 1
vps_ceiling 64 128

After changing these parameters, we are getting this error.

kgpspawn failed:category = 27143, depinfo = 24, op = pipe, loc = skgpspawn2


Please advise.

[Steven E. Protter]

Shalom,

The error seems vaguely familiar.

You don't state how much memory your system has, is possible you have exceeded capacity with one or more parms.

Also, take out the formulas. You don't nee them and they can cause errors themselves. a

Lastly to diagnose:
Change everything back and change fewer parameters to isolate which one caused the problem.

SEP

[moodi]

Thanks steven for quick reply, I have revert back parameters but problem is, machine is in production and we can change parameter on production machine, becasue after changing the parameters we need to reboot machine.

Please help me which parameters are required to change.

Thanks

[Bill Hassell]

Several parameters need adjusting and as Steven said, dump the formula entries, at least for the most common items:

> dbc_max_pct must be changed to a percentage of your RAM that works out to about 800-1600 megs. Since memory size changes (add more RAM), it's much simpler to just specify bufpages=250000 (250k) for 1000megs of 4k pages. When bufpages has a non-zero value, dbc_min_pct and dbc_max_pct are ignored (which is what is desired)

> maxuprc is set using a formula. This is the limit of all running processes owned by a single user. It should be a fixed number (like 3000) rather bouncing around with nproc. You can monitor process counts by user with this command line:

ps -eo user= | sort | uniq -c | sort -n

The owner(s) with the largest number of processes will be shown at the bottom of the list.

> ninode is more than 10x too large. Change it to 2048 and forget it. This parameter is primarily for in-core inodes from the /stand filesystem.

> maxusers is not a kernel parameter at all but is used to adjust other parameters. In some cases, it is reasonable but others, it is way off, usually making a parameter far too large an wasting kernel memory.

> nproc set to 8000 may be reasonable but check the actual usage with sar -v 1

> nfile to be set to 63000 may also be reasonable but check with sar -v 1. In both cases, when the usage approaches 75%, schedule a reboot so you can double the value.

[arijit dey]

What is memory for the above configuration!

Is it for 12 GB RAM ?

What is the swap memory size?

Please clarify so that somebody else can use this config.

[Steven E. Protter]

My recommendation to dump the formulas was straight from Bill Hassell's Advanced Systems Administration Class from 2004.

Why is a thread that's been idle being openned after 12 days?

If the owner wishes to pursue it, he'll provide more information. I suspect this is resolved and should be left alone until the author closes it or says otherwise.

ITRC Ettiquette, lets not clutter the forum with inactive threads.

SEP