On Monday 07 July 2008 20:14:04 Gabriele Modena wrote: > On Sat, Jul 5, 2008 at 4:14 PM, Henrik Austad <henrikau@xxxxxxxxxxxxxx> > wrote: > > [...] > > I also worked on a real-time project involved in benchmarking the > rt-preempt patchest > and to determine the behaviour in scheduling real-time IRQ threads > with the CFS scheduler > (roughly speaking, I was trying to obtain Quality of Service by > scheduling interrupt threads). > > After my bachelor I did not have to time to further work on it. I > applied for Summer of Cod, but did not go well :) > I'm still intentioned in carrying the work on as a personal project > starting in the next few weeks though. I'm a bit bogged down at the moment, so I won't start working full-time on this before mid-august. Right now, I'm reading through UTLK, kernelnewbies, lkml and various web-resources trying to put the magic pieces togheter :-) > > I'm going to look at implementing an EDF scheduler, with the rt- patch > > series, ands finally try to measure the performance and latency compared > > to other schedulers (O(1) and CFS) > > Are you going to implement it as a CFS's scheduling class? I'm curious > about any sort of overhead that could be encountered in such a case. To be honest, I don't know yet, that's part of my project. :-) It would be interesting to rip out the entire scheduler and implement an EDF from scratch, but given the nature of EDF-scheduling, a lot of extra code must be changed as well (since a process must notify the kernel wether or not it's aperiodic, expected runtime, frequency) > I would also be curious to benchmark the rt-groups patchest (that IIRC > should allow the possibily to schedule by using EDF and control > groups). Hmm, this is actually new to me. How does it schedule a group? And how is processes/threads scheduled within that group? -- mvh Henrik Austad
Attachment:
signature.asc
Description: This is a digitally signed message part.
