This is more of a place holder for a problem discussed here:
http://lkml.org/lkml/2005/5/7/98
I am sending out this patch as a record of the load balance problem that
exists with NO_IDLE_HZ implementations.
Ingo had mentioned at OLS that it would be nice for any
non-idle-cpu-waking-up-sleeping-cpu.patch to maintain the load
balancing algorithms intact. Basically the algorithm is offloaded
from a idle cpu to non-idle CPU and potentially lets the idle CPU
sleep for longer periods.
I have some ideas to guarantee that load balance algorithms are not
changed and yet let idle CPUs sleep as long as they want. But I have
been unable to implement them because of other pressing (read dyn-tick!)
work. Plan to work on such a patch and send out for discussions (& flames :)
as soon as possible.
--
Thanks and Regards,
Srivatsa Vaddagiri,
Linux Technology Center,
IBM Software Labs,
Bangalore, INDIA - 560017