From: george a. <ge...@mv...> - 2001-09-06 20:40:58
|
You might want to take a look at: http://sourceforge.net/projects/rtsched/ Probably more than you want, but... Seperating the queues shortens the goodness function (it need not check rt tasks) and should be a win, assuming you don't try to use the same goodness function on both. George Hubertus Franke wrote: > > * Mike Kravetz Wrote > > > > > > I'm in the process of rewriting the MQ patch in the hope that > > it will be more 'Linus friendly'. One thing I want to do is > > keep as much code as possible common between the UP and SMP > > versions of the scheduler. In the MQ scheduler, we keep a > > separate real-time runqueue. I was wondering if it would make > > sense to keep a separate real-time queue in the UP case. I > > know several people have played with the idea of a separate > > real-time queue in the past. Does anyone recall the outcome > > of this work? > > > > Mike, these are the issues that will arise. > (a) if you integrate with the current MQ structure > the RT-Q will be on its own cache line, so you would have > in general one extra cacheline access and test, whether you > have RT tasks or not. > (b) during add_from_runqueue() and del_from_runqueue() you will > still have to distinguish which queue the tasks fall on. > (c) on the plus side, if you have RT tasks you can probably serve them > quicker then running through the whole single queue. > > So all in all, one will encounter a few more instructions under (b) and > and additional cache hit under (a). > > > In addition, does anyone know of a benchmark that can be used > > to test 'real-time performance' with respect to the scheduler? > > I want to make sure we don't degrade the real-time case. > > > > Thanks, > > -- > > Mike Kravetz kr...@us... > > IBM Linux Technology Center (we're not at Sequent anymore) > > > > _______________________________________________ > > Lse-tech mailing list > > Lse...@li... > > https://lists.sourceforge.net/lists/listinfo/lse-tech > > > > > > _______________________________________________ > Lse-tech mailing list > Lse...@li... > https://lists.sourceforge.net/lists/listinfo/lse-tech |