[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: max # of processes in debian testing

On Tuesday 12 February 2002 17:53, Joey Hess wrote:
> debian@clouded.org wrote:
> > Hmm. That's great for shell scripts and interactive shells, but what
> > about, for example, the number of sendmail children on a box? My problem
> > is I have a machine that can handle more than 300 sendmail children at
> > any given time, but once it hits 300 I've got to unplug the network and
> > wait for those children to die before I can fork new processes, such as
> > ps or kill.
> >
> > Would fixing this be as simple as sticking a ulimit command in my
> > sendmail init script or does sendmail fork off into it's own environment
> > making that useless?
> No, ulimit will not help you because linux has a hard limit on the
> number of processes built into the kernel.
> http://linuxperf.nl.linux.org/general/kerneltuning.html
this is something that seems to have changed with the new scheduler from Ingo 
Molnar; I had no problems running more than 1000 processes as a normal user 
on my 2.4.18-pre8-mjc kernel, and I can push my process ulimit up to 5119. 
I've no idea how exactly the kernel determines these ulimits though,  since 
they greatly vary on different  machines, on auric for instance I'm able to 
feed "ulimit -u" arbitrarily high values, such as 100 000 000, although I 
guess it's rather unlikely that auric would be able to handle such an amount 
of processes ;-) 
So has anyone an idea of how exactly the kernel derives these limits?

Yven J. Leist - leist@beldesign.de

Reply to: