1834149 Members
2776 Online
110064 Solutions
New Discussion

Re: nkthread

 
SOLVED
Go to solution
khilari
Regular Advisor

nkthread

Hi people, i was was cmruncl to get my two node cluster back up, when it gave me the error that nkthread size is full. It is 499 and 499 used. Now there is a formula which i really dont want to mess with. So, how should i go about changing or increasing the value..
And secondly, i thought nkthread was java related so why is it interfering with re forming the cluster.
Thanks
3 REPLIES 3
Prashant Zanwar_4
Respected Contributor

Re: nkthread

This has happened at my end also.. and you have to bump system parameters for this to be avoided. You need to adjust your kernel values.. Check below paper and do it accordingly..
If system is new, you need to change it mostly..
Check what type things have changed on system..

Hope it helps
Prashant
"Intellect distinguishes between the possible and the impossible; reason distinguishes between the sensible and the senseless. Even the possible can be senseless."
Steven E. Protter
Exalted Contributor

Re: nkthread

Shalom,

Take out the formula.

As a matter of course, many admins take out these formulas which limit flexibility and often serve no purpose other than to drive admins up a wall.

I don't think nkthread is java releated though if its not high enough java may not run right.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
John Bigg
Esteemed Contributor
Solution

Re: nkthread

nkthread is only java related in that java can take a lot of threads so typically needs to be increased in java environments. Any system which uses a lot of kernel threads may need to have this parameter increased. Note that every process takes at least 1 thread and processes such as java and Serviceguard's cmcld can take more.

If you are unconfortable removing the formula you could consider modifying it to add a larger static value to handle the extra threads taken by multi threaded processes.

i.e. I think the default fomrula is:

((nproc*2)+16)

you could change this to:

((nproc*2)+256)

so it will still be automatically scaled if nproc is increased but will have an additional 240 threads on top of the old value. Or maybe change the multiplier if you think this would be more appropriate for your environment.