Digg - What is the Completely Fair Scheduler? (via Mowser)

Introducing Digg Dialogg!
Check out the first Digg Dialogg with Nancy Pelosi. More guests to be announced soon!

What is the Completely Fair Scheduler?

immike.net — A pretty detailed explanation about whats up with the upcoming 2.6.23 Linux kernel release.

Bury

Favorite?

bedoyle 1 year 28 days ago, made popular 1 year 27 days ago

471 diggs digg it

49 Comments Who Dugg It? Blog It Email It

« View All Comments

serend, on 10/10/2007, -4/+3I still think that CK's Staircase / Deadline scheduler was and is better.. Too bad CK had to go out on a limb and cause some drama.

YourDoom123, on 10/10/2007, -2/+2isn't Deadline an IO scheduler? this is talking about the cpu scheduler which was never revealed in the kernel options menu: there's only one.

Urusai, on 10/10/2007, -0/+3The difference between the CPU and IO schedulers isn't hardly covered. Really, the IO scheduler is what sucks in Linux. Try copying a big file in the background and watch your desktop stutter and freeze.

DrDabbles, on 10/10/2007, -1/+4If you don't know what you're talking about, don't speak. To say it's the IO scheduler is to do a disservice to the topic. The IO sched. needs more work, though CFS does a good enough job as does deadline, but the IO scheduler is showing age. The CPU scheduler known as CFS tries to simplify scheduling, thus lowering overhead and theoretically increasing interactivity and process responsiveness without sacrificing actual processing performance.

The problem CK has always had with his patches is that he thinks one way and starts off running. When his staircase scheduler first came out (This predates the 2.6 kernel...I tested it in embedded network devices I was developing at the time), it sucked BADLY for a system under any real load. It worked great for a desktop running a web browser, but do any actual processing and watch out. His other patches introduce stability issues and break compatibility in many cases. This is also why resiser4 has not been adopted to mainline vanilla kernel.

For a good dose of reality, and a history lesson...go read kerneltrap XVampireX, on 10/10/2007, -0/+1DrDabbles is right, It's a great thing to have at least a few people who are less into hacking the kernel but more into how it works :)
I'm a big fan of kerneltrap, read it every week, and sometimes even several days a week.

shethinkmefunny, on 10/10/2007, -0/+1I have yet to play with either scheduler, but I've tried a few kernels with the -ck patches, and while it makes standard usage seem snappier, it has a few show stopping trade-offs, specifically multitasking during large file transfers. With the standard kernel, transferring 100 GB of data from one drive to another (separate SATA drives on different channels in my case) will slow the system somewhat, but not enough to be a real hindrance. With the -ck patched kernel, that same 100 GB transfer makes all my running applications cease to respond, albeit with up to 15% faster file transfer speeds. It's irritating, to say the least, to see my hyperthreaded 3.4GHz machine with 2 gigs of RAM all but lock up during a large file transfer.

If his coding style carried over to his scheduler, I would be more inclined to go with the CFS, as everything I've read suggests that while it's not particularly stellar in any one area, it's very competent across a very broad spectrum of usage scenarios.

XVampireX, on 10/10/2007, -0/+1IO Scheduling always needs to be better. Also what file system you choose makes a big difference on the speed of your file transfers and on how much it affects the desktop performance.

Login or register to add a comment

Create a new account or login to join in the conversation on Digg. You'll also be able to Digg stories to help promote things you like.