Flemming Danielsen wrote:
> I have found in the past that it is frustrating to find resources
> rctl that is not in sync between projects and zones definitions and
> as a costumer I would like to see them in sync.
> It would also enable me to isolate applications in zones that tend to
> blow up the zone where I have other part of the workload running
> (other projects).
I have added a project.max-processes resource control to the prototype
so you can now also limit the number of processes per project inside a
zone. As suggested by Steve, project 0 in the global zone is exempt from
the project.max-processes rctl.
Would there be any interest in having BFU archives of the prototype
available for people to experiment with?
> I would like you to consider something more appropriate then a
> unlimited value of processes in global zone. We did an experiment
> where we created about 400-500 zones and we to hit some limits in the
> default kernel when we reached 80000 lwps. This will also affect the
> process limits. I expect there is a kernel structure that needs to
> be tuned, but I have not had the need to build a 500 zones system
> yet :-) But since you put a lower limit you might investigate how o
> put in a upper limit (based on kernel structures)
I don't think an upper limit on the the number of processes in the
global zone is the right way to handle that failure mode. The failure is
presumably not caused by running out of processes slots, but most likely
due to lack of space for LWP stacks. And since there is no direct
relation between the number of processes and the number of LWPs, I think
that the max-processes rctl just isn't the right tool to solve this
Menno Lageman - Sun Microsystems - http://blogs.sun.com/menno
zones-discuss mailing list