Is The Linux Kernel Scheduler Worse Than People Realize?

[GreatEmerald]

But how does that affect power consumption? If all work is going on one core, the other cores can go into deeper sleep.

[jacob]

Glad to see attention brought to this matter, it really is a problem... And it might not be that schedulers on other systems are better, it's just that ours isn't good enough.

You know there's something wrong when I get better results with noop than cfq, yet cfq is always the recommended default everywhere because it is theoretically better for some things. I wouldn't exactly call noop satisfying either. People generally praise BFQ but again that is geared towards unicores...

I mean I don't know the technical details of any of this, I just know it's a mess...

NOOP and CFQ are IO schedulers. This article is about CPU schedulers, which is an entirely different matter.

[jacob]

But how does that affect power consumption? If all work is going on one core, the other cores can go into deeper sleep.

It's not as simple, because if one core is doing all the works, it may need to crank up its frequency. Still, this is basically what ARM's HI-LO chips do.

[Guest]

Is there a way to possibly test out manual pinning with various applications and games (or in other words, what's the program or command)?

Also, just to make sure I understand the difference between an I/O and CPU scheduler; CFQ, Noop, Deadline, and BFQ are I/O schedulers, and BFS is a CPU scheduler? What CPU schedulers exist, and can you easily switch between them like you can with I/O schedulers? Doing a 'dmesg | grep scheduler' only showed me what I/O schedulers I have.

Edit: Doing a bit of research, it looks like Completely Fair Scheduler (CFS) is the default CPU scheduler, but I don't really see any other options available aside from BFS.

[VikingGe]

Desktop kernels should use BFS anyways.

I've recently switched back to CFS from BFS because CFS actually performs better these days, at least for my typical workloads. Core hopping is a serious problem with BFS (under medium load conditions, single-threaded tasks perform like shit because they are being moved to another core every 10ms) - which is almost non-existent with CFS (running a six-core Phenom II here), and using Firefox while compiling something with all cores busy is also not a problem anymore. BFS was great a few years ago when CFS caused serious latency issues under heavy load conditions.

[liam]

Oh, there is an upcoming Linux scheduler? I didn't know that …

http://lwn.net/Articles/682391/

Incidentally, I'd imagine most people found out about this paper from here: http://lwn.net/Articles/683744/

[liam]

@Espionage724
Cpuset

Also, you can run htop with admin privileges and pin tasks using a tui.

[Alejandro Nova]

NOOP and CFQ are IO schedulers. This article is about CPU schedulers, which is an entirely different matter.

I see a real problem here. BFS isn't markedly better than CFS, and Con Kolivas' relationship with the kernel maintainers is cold. BFQ, OTOH, is almost always better than CFQ, but I think one of the reasons why BFQ doesn't see a warmer reception comes, precisely, from being a) confused with or b) associated with BFS. Also, both tend to appear in the same patchsets.

[jayaura]

I've had some bad experience with buggy software quite a lot of times. Something misbehaves and HDD activity shoots up. Then everything freezes. Sometimes things get back to normal, sometimes I end up restarting the system. I never expected this to happen in a linux machine. I guess this is the problem with the I/O scheduler ?

[Guest]

I've had some bad experience with buggy software quite a lot of times. Something misbehaves and HDD activity shoots up. Then everything freezes. Sometimes things get back to normal, sometimes I end up restarting the system. I never expected this to happen in a linux machine. I guess this is the problem with the I/O scheduler ?

What's the default scheduler (can use "dmesg | grep scheduler" to find out)? If it's noop or deadline (Ubuntu defaults to deadline), then (from what I've seen), anything that uses the HDD pretty hard can cause other things to freeze up for a bit, including the mouse