Re: [PATCHSET] Linux 2.4.19-pre7-jam1

[Dieter Nützel]

On Friday 19 April 2002 01:44, Robert Love wrote:
> On Thu, 2002-04-18 at 19:36, Dieter Nützel wrote:
> > No uptodate O(1) patch for 2.4. Very sad.
> > So there isn't any change to see a current preemption patch on top of
> > vm33 and O(1).
>
> I am working on backports of all the O(1) scheduler changes in 2.5, the
> pending changes, and some other misc. bits.  I also have versions of the
> migration_thread and affinity stuff for 2.4.

Ahhh, I'm very happy to hear this. GREAT!!!

> I will release a general O(1) patch and a patch for -ac soon - hopefully
> tomorrow or Monday.  I have no idea if it fixes the problems you are
> seeing, because I have no idea what caused a regression in the O(1)
> code.

I couldn't figure it right, but I've rechecked simple 2.4.18 without all 
"pending stuff" and it shows the same numbers as 2.4.19-pre7+AA.

When I apply Ingo's originall sched-O1-2.4.18-pre8-K3.patch or J.A.'s the 
latency goes bad.

> > No, lowlatency didn't come close to preemption+lock-break (best latency
> > numbers for 2.4.17-preX-rml, were ~2.9ms max).
>
> Good to hear ;)
>
> > I'm under the impression that "all" development is focused on 2.5.x, now.
> > Even the VM stuff show no mayor growth ;-(
>
> That is the point of 2.5 :)

Yes, I know, it is more fun...;-)

> Development => !Stability and people need to start using 2.4 to get work
> done, not reap faster and faster benchmarks times.  I seriously suspect
> 2.4 is performing fine right now for what you are doing, anyhow.

Nope, it is my devel machine and I do "heavy" C++ VIS app and XFree86 DRI 
compiler runs in the background during browsing, mail, etc. --- "Normal" 
workstation use...;-)

My dual Athlon MP/XP is some days around the corner ;-(

> Also, a lot of VM work is happening in 2.4 (and not in 2.5 even, at the
> moment).  2.4.19-pre has seen a few of the -aa bits merged and should
> see most of the others in due time.
>
> There is also Rik's -rmap for 2.4 ...

Wouldn't start a VM flameware, again but -rmap didn't come close in VM 
throughput. 2.4.17-preX-aa-O(1)-rml was the killer.

Thank you for all your work!

-Dieter

BTW As always, send me copies, please ;-)