[linux-audio-user] 2.6.13 kernel: which patches for better latency

[Reuben Martin]

On 9/12/05, Reuben Martin <reuben.m at gmail.com> wrote:
> On 9/12/05, Eric Dantan Rzewnicki <rzewnickie at rfa.org> wrote:
> > On Thu, Sep 08, 2005 at 08:38:02PM -0500, Reuben Martin wrote:
> > > On 9/7/05, guy <sayhi2guy at tiscali.co.uk> wrote:
> > > > Hi
> > > > I'm just about to compile a 2.6.13 kernel for low latency audio work.
> > > > The docs I have accessed are either confusing (sorry if its just me ;0)
> > > > ) or rendered out of date by kernel developments.
> > > > I know that some features such as pre-empt are now in the kernel.  Which
> > > > patches (if any) would you folks advise that I should apply.  I am also
> > > > unsure if should enable the 'big kernel lock' feature.
> > > > Thanks in advance for any help
> > > realtime-preempt patches are the latest and greatest for latency
> > > improvements. They are very actively maintained by Ingo Molnar, same
> > > guy who created the preempt patches for the 2.4 kernel series.
> >
> > For historical clarification ( and someone correct me if I'm wrong on
> > any of this ), Ingo did a set of low latency patches for 2.2/2.4 that
> > basically showed that breaking up long code pathes could improve linux's
> > latency performance. These were proof of concept patches, in a sense,
> > but not accepted, nor intended, for the mainline kernel.
> >
> > Andrew Morton created a smaller and more tightly focused set of
> > long-code-path-splitting low-latency patches for 2.4 inspired by Ingo's
> > earlier work. These were intended to find an approach to achieving
> > low-latency more acceptable to the kernel devs.
> >
> > Robert Love created a set of Preemption patches for 2.4 that were
> > commonly applied along with Andrew's LL patches. The preemption patches
> > were a different approach to achieve low-latency which allowed code
> > paths which might run for a long time to be marked as pre-emptible.
> >
> 
> I know all three of them contributed. I'm not sure who's was merged
> into 2.5. I remember I used to use a combination of two diffferent
> latency patch sets.
> 
> 
> > Much of Andrew and Robert's work on 2.4 was incorporated into the 2.5
> > development kernel and thus the 2.6 series. However there were still
> > latency problems with 2.6, so Ingo has again taken the lead in producing
> > the current series of realtime-preemption patches. A lot of this work is
> > now present in the mainline 2.6.13.x kernel.
> 
> If I understand correctly, chunks of his patch are slowly absorbed
> into the mainline a bit at a time while he continues to keep finding
> ways to shorten the length of code paths.
>