Ingo Molnar Tests New BF Scheduler

Last time this issue arose (two years ago? three years ago?) Con Kolivas had been so frustrated with the decisions by amongst others Linus that he quit kernel development and didn't want to have anything to do with it anymore. Also, Linus and probably Ingo Molnar also were a bit fed up with Con's allegations that Linux didn't have good 'responsiveness' for the desktop.

Ingo Molnar just told it was not 'measurable' in a significant way that Con's Rotating Staircase Deadline Scheduler was better on the desktop than what was currently in Linux.

I have tested those -ck patches back then, including RSDL Sched and it felt 'more responsive'. Placebo or not, I still don't know. Anyway, it's good to see Ingo Molnar is seriously looking into this issue now. At least, to some spectator such as I that seems to be the case. And moreover, it shows maturity not to tell each other to piss off after some fruitless debates.

BTW BrainF*ck scheduler, what would that look like? +++]]--[[> or so?

Oh, this is a nice explanation of the name by the man himself:

Quoting:

Why "Brain F*ck"?

...Because it's designed in such a way that mainline would never be interested in adopting it, which is how I like it. ... Because it throws out the philosophy that one scheduler fits all and shows that you can do a -lot- better with a scheduler designed for a particular purpose. I don't want to use a steamroller to crack nuts. ... Because I must be fucked in the head to be working on this again.

http://ck.kolivas.org/patches/bfs/bfs-faq.txt

A bit strange when Con tells this is for "flash video's on slow systems" that Mr. Molnar uses an 8 CPU system (!) to do the benchmarks, thereby confirming what Con is saying.

Hans -

I remember when that brouhaha was going on and I SOOO wanted to reach through the wires and strangle Ingo and others.

I don't say that as any kind of kernel developers, but...

I did spend 10 years as a performance geek and capacity planner for large mainframe centers, and can well appreciate both the importance of responsive systems and the difficulty of getting a good measurable handle on them.

Methinks a few kernel developers would do well to pull all those beans out of their ears.

Hmm, it seems like Ingo & co have always been more concerned about large-system performance then about desktop (single pentium II etc.) performance, but I may be wrong?

hans -

No, you are right. Wasn't so bad in the old days -- back before we were running so much audio and video and whatnot, but that ain't today's world.

Servers matter, but desktops outnumber them. By lots.

And when it comes to desktops, I'm inclined to think that what it feels like matters almost more than what is measurable.

Quoting:I'm inclined to think that what it feels like matters almost more than what is measurable.

The feel does matter more than the measurements. (And I don't mean that in a TOS-liable manner!)

The wide variety of hardware configurations and capabilities makes any such measurements essentially useless. Until someone can isolate the performance impacts of the display, the CPU(s), the FSB, the memory subsystem, and whatnot else, the most valid testing will be a double-blind survey.

gus3 -

You might be surprised by the meaningful results you can get without taking every last bit of hardware into account. The task comes down to measuring things the right way --- response distributions instead of averages, or delivered throughput vs. theoretical charted over bunches of timeslices, etc. for things like audio/video playback.

You can't exactly get it, but you can find major markers that line up well with subjective experiences.

Quoting:Hmm, it seems like Ingo & co have always been more concerned about large-system performance then about desktop (single pentium II etc.) performance, but I may be wrong?

You're absolutely right. With the popularity of netbooks and nettops, which is still growing by leaps and bounds, even in business and with the economic crisis forcing businesses to keep older machines in service longer the need to tweak performance to the max for the low end is very real right now.