Re: [PATCH 2 of 4] Introduce i386 fibril scheduling

[Ingo Molnar <mingo@elte.hu>]

* Mark Lord <lkml@rtr.ca> wrote:

> >also, we context-switch kernel threads in 350 nsecs on current 
> >hardware and the -rt kernel is certainly happy with that and runs all 
> >hardirqs
> 
> Ingo, how relevant is that "350 nsecs on current hardware" claim?
> 
> I don't mean that in a bad way, but my own experience suggests that 
> most people doing real hard RT (or tight soft RT) are not doing it on 
> x86 architectures.  But rather on lowly 1GHz (or less) ARM based 
> processors and the like.

it's not relevant to those embedded boards, but it's relevant to the AIO 
discussion, which centers around performance.

> For RT issues, those are the platforms I care more about, as those are 
> the ones that get embedded into real-time devices.

yeah. Nevertheless if you want to use -rt on your desktop (under Fedora 
4/5/6) you can track an rpmized+distroized full kernel package quite 
easily, via 3 easy commands:

   cd /etc/yum.repos.d
   wget http://people.redhat.com/~mingo/realtime-preempt/rt.repo

   yum install kernel-rt.x86_64   # on x86_64
   yum install kernel-rt          # on i686

which is closely tracking latest upstream -git. (for example, the 
current kernel-rt-2.6.20-rc7.1.rt3.0109.i686.rpm is based on 
2.6.20-rc7-git1, so if you want to run a kernel rpm that has all of 
Linus' latest commits from yesterday, this might be for you.)

it's rumored to be a quite smooth kernel ;-) So in this sense, because 
this also runs on all my testboxes by default, it matters on modern 
hardware too, at least to me. Today's commodity hardware is tomorrow's 
embedded hardware. If a kernel is good on today's colorful desktop 
hardware then it will be perfect for tomorrow's embedded hardware.

	Ingo