Ingo Molnar: BFS vs. mainline scheduler benchmarks and measurements
thread.gmane.org
thread.gmane.org
As I software engineer, I realize that making different schedulers for different architectures is "inefficient" (unless development/test resources are increased). But if somebody is willing to put in the time and effort to make me a scheduler that works best on my hardware, I'll gladly take it.
I have to agree with Con's remarks here. Seems like an underhanded move by Ingo to give this thing a bad reputation (and hence kill developer and beta tester interest) before it comes out. And judging by the comments here, he may have succeeded. Looks like open source has lots of politics too. :)
/me sees Ingo run off to find the right combination of hardware and benchmark to prove his point.
[snip lots of bullshit meaningless benchmarks showing how great cfs is and/or how bad bfs is, along with telling people they should use these artificial benchmarks to determine how good it is, demonstrating yet again why benchmarks fail the desktop]
So, to paraphrase: BFS is vastly superior, but not in any way that you can actually measure.
Con made a valuable contribution back when the choice was between staircase-deadline and the old mainline O(1) scheduler, but he's since turned himself into a crank.
Ingo benched this on a two-socket, 8-CPU system. CK made it blatantly obvious he was targeting single-socket multi-core systems, or UMA SMP systems. Further, he's seems pretty adamant about not tweaking the scheduler for anything beyond single-socket multicore or UMA SMP, to say nothing of more exotic CPU/Memory architectures.
Writing a scheduler is not for the faint of heart.
What bothers me about this whole affair is that the egos of both parties seem to preclude them from working together to get to a 'best of breed' situation.
Both Con & Ingo have a huge NIH problem.
> Building a scheduler that doesn't scale to an 8-CPU
> system seems like madness to me, given the very
> predictable trends in desktop hardware.
Which is exactly the kind of machine that I'm sitting in front of at this very moment and have owned for more than two years. I realize this puts me on the leading edge, but any scheduler that can't cope with this in a reasonable manner is a bit of a joke.
He did it again on a single-socket 4-CPU system with HT. He got relatively the same results, including the curious marginal win for BFS on kbuild -jN where N == the # of logical processors.
From my understanding, your interpretation is not correct. BFS is about latency, not throughput. Ingo chose all the wrong benchmarks.
Also, it's a dick move by Ingo. CK hasn't even made a public announcement about BFS yet.
Finally, you should use code that has technical merit, regardless of the personality behind the writer. I imagine the kernel you're running on has code in it written by some pretty unfriendly people.
Then how about CK showing up the right ones?
Also, it's a dick move by Ingo. CK hasn't even made a public announcement about BFS yet.
He announced it publicly enough that I knew about it, and I haven't followed this stuff in years. Given that, I hardly think that commenting on his efforts on LKML constitutes a premature "outing".
Finally, you should use code that has technical merit, regardless of the personality behind the writer. I imagine the kernel you're running on has code in it written by some pretty unfriendly people.
Indeed: one of them murdered his wife. But you brought up personalities, not me. The part of CK's post that I quoted is rather dickish, but I deliberately avoided commenting on that.
Why? It's not as if CK is trying to get it into the mainline. I think the only reason why CK got angry is that Ingo's benchmarks appear to be deliberately chosen to make BFS look bad. Anyone who took the time to read the documentation would realize that testing throughput is not correct. That'd be like benchmarking Python and C using matrix multiplication and concluding that Python sucks. Two different tools for different purposes.
He announced it publicly enough that I knew about it, and I haven't followed this stuff in years. Given that, I hardly think that commenting on his efforts on LKML constitutes a premature "outing.".
I think the announcement was mostly the work of Reddit et al., but fair enough.
Indeed: one of them murdered his wife. But you brought up personalities, not me.
Reiser was a crazy one. ;) I actually only brought up personalities because I was under the impression that you wouldn't try CK's patch because of his personality. I say if there's interesting code out there, it shouldn't go to waste because of the quirks of the author. I apologize if I misinterpreted.
What does one have to do with the other? If I'm going to spend the necessary 20 minutes to compile a new kernel using CK's patch, I want some evidence that I'm going to be gaining something.
Ok, I can accept that. For me personally though, the quality of -ck patchset is enough for me to give CK's new stuff a shot.
Accusations like "/me sees Ingo run off to find the right combination of hardware and benchmark to prove his point." is far from the level of civility I would entrust my favorite OS's kernel. The proper response would be something in the line "it will suck on this benchmarks, but it will excell in these others and in these hardware configurations" and show numbers. It's always possible to act in a civilized way.
That said, CK seems only a bit off when compared to a couple others. There are a lot of people who like to be quite caustic from time to time. I must confess I have to refrain myself frequently and HN is not LKML.
And, more important: this whole discussion is more or less pointless. It doesn't matter how good a modern scheduler is, all of them impose a very little penalty on a modern kernel. Are we really going to fight this bitterly over a couple percentage points?
Would that be so hard as to include a "-scheduler=" switch in the kernel for it to be selectable at boot time (and so I can pick one that fits both my netbook and my SGI 4096 CPU monster when I want to)? This would pretty much end this discussion.
Then perhaps you should pay Con Kolivas to prove the value of his freely provided no-warranty work-in-progress code for you.
I have also published a great deal of open source software, much of it successful, and some of it almost universally used. Unfortunately, that does absolutely nothing to bolster either of our arguments.
If you published an essay containing controversial opinions, would you demand that it be immune to criticism on the basis that you're giving it away for free
You asked for: "I want some evidence that I'm going to be gaining something".
If we try to apply that to your analogy, why is it the author's responsibility to not only produce the work, document the work, but then also supply their own literary criticism so that you can decide if the work is worthwhile, all before the essay is completed and ready for publication?
Open source developers regularly produce software just because they want to, and feel no impetus or desire to convince you of anything. That's not, as you put it, some sort of "narcissistic" behavior, but rather a desire to simply share without additional burden or demands by others.
BFS was dramatically worse across the board in all the benchmarks, except curiously when the # of processes == # of logical processors -- to me that's evidence of extreme naivete in its design.
If I were to assign blame for the latency problems I experience, I would place it on the 100-call-deep stacks that are used to deliver events in Gtk+ and Qt, and on the binary Nvidia driver's lack of good 2D+compositing acceleration.
I would be pretty worried with an anesthesiologist that disregarded his measurements... Either you can measure or it doesn't exist. If it exists and you don't measure, you'd better write something to measure it.
Perhaps you could time clicks to open new tabs in your browser. Or switch virtual spaces.
With the background work, you should notice no or very little difference. Since the interactive work should get highest priority(this is what you are aiming for).
You can try these tests without bfs patches too. You can see how the default kernel/X11/driver is not optimal - and how things can be improved.
It's still much nicer than OSX, and windows however.
BFS had been ported to one of the Custom Roms for Android and has already shown increased speed improvements.