On Thu, Jun 25, 2015 at 09:48:55PM +0100, Mel Gorman wrote: > On Wed, Jun 24, 2015 at 05:50:28PM -0500, Nathan Zimmer wrote: > > My apologies for taking so long to get back to this. > > > > I think I did locate two potential sources of slowdown. > > One is the set_cpus_allowed_ptr as I have noted previously. > > However I only notice that on the very largest boxes. > > I did cobble together a patch that seems to help. > > > > If you are using kthread_create_on_node(), is it even necessary to call > set_cpus_allowed_ptr() at all? > Yup kthread_create_on_node unconditionanly calls set_cpus_allowed_ptr(task, cpu_all_mask); It does it to avoid inherting kthreadd's properties. Not being familiar with scheduling code I assumed I missed something. However it sounds like it should respect the choice. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>