2007-10-23 20:00:26

by Jan Glauber

[permalink] [raw]
Subject: [RFC PATCH 0/2] use virtual time for CFS on s390

Hi Ingo,

an outcome from the previous discussion about a virtual sched_clock() on s390
was that scheduler_tick() should also be called based on virtual time.
The second patch changes the scheduler_tick() call to only happen after a tick
passed for the virtual cpu.

The patches cause nothing obvious to break, numbers from top look sane but
I've seen something strange...

For a simple make -j6 workload top reports processes very often to be in
state <defunct>. Thats' not terribly wrong since this can happens also
without the patches but I wonder if this indicates that the scheduler behaves
badly and does less often schedule the parent processes that gathers the
waiting zombie processes?

Are there any indicators in the sched_debug output that I could look for?

Jan