Messages in this thread | | | Date | Mon, 23 Jun 2008 20:15:04 -0400 (EDT) | From | Steven Rostedt <> | Subject | Re: [PATCH 0/3] RT: scheduler newidle enhancements |
| |
On Mon, 23 Jun 2008, Gregory Haskins wrote:
> > Hi Ingo, Steven, Thomas, > The following series applies to 25.4-rt6 and is "ready for inclusion" from > my perspective. However, due to their nature I am sure we will want to get > some review feedback before being considered for acceptance. > > These patches are probably suitable for consideration for > mainline/sched-devel as well, but PREEMPT_RT gains the biggest boost from > them since it tends to context switch much more frequently than mainline. > > This series makes some adjustments to the way we do newidle balancing. > These tweaks were discovered while doing some related scheduler R&D that is > not quite ready for the light of day. But these patches do offer a > substantial boost (6-12%) in network performance. They may help in other > areas as well, but networking is where our focus remains currently. > > I think the patch 1/3 is a good one, but it requires careful review. Patch > 2/3 is based upon some work I had seen PeterZ submit, so its possible to > just pull his work (if it hasnt been already) as opposed to this patch, if > desired. Lastly, Patch 3/3 is a bit controversial so I put it last for > easier cherry picking of the first two. It does help in our testing, but it > will have to be reviewed carefully. > > Comments/feedback/bugfixes more than welcome.
Greg,
I don't see anything wrong with these patches. If either Ingo or Peter have any issues, I hope they speak up soon. I'll queue them all for 25-rt8 (after I get rt7 out the door).
-- Steve
| |