Digg - CFS scheduler to appear in Linux kernel 2.6.23 (via Mowser)

Call for questions
Submit and vote up questions you'd like to see answered by Kevin & Jay at the next Digg Townhall on 11/18.

CFS scheduler to appear in Linux kernel 2.6.23

linuxinsight.com — The Linux kernel process scheduler, as you know it, has been completely ripped out and replaced with a completely new one called Completely Fair Scheduler (CFS). How fair it will be, remains to be seen, here's what its original creator Ingo Molnar says: CFS basically models an "ideal, precise multi-tasking CPU" on real hardware.

Bury

zlatko 1 year 131 days ago, made popular 1 year 130 days ago

421 diggs digg it

48 Comments Who Dugg It?

« View All Comments

russellnation, on 10/11/2007, -0/+5being new to linux, i have no idea what this means. but is sounds cool

zlatko, on 10/11/2007, -1/+9Process scheduler is really a heart of any multitasking OS. Your processor (let's assume a single-core one for this explanation, ok?) can actually do just one thing at a time, and scheduler's task is to constantly switch runnable tasks on the CPU and thus create the illusion of multitasking. It tries heavily to be fair and efficient (not lose more time scheduling than doing the real work). So, scheduler is actually not a Linux specific thing.

What is interesting with this CFS scheduler is that it completely replaces the previous one, so 2.6.23 will, in a way, be a completely new beast. Something like: you've been driving your trusty old Ford for years, and then suddenly you decide to rip its engine out and put a Honda engine inside. :)

geminitojanus, on 10/11/2007, -0/+12It's also important to note that the new scheduler is simply an optimization of the previous scheduler ideals (it still runs in O(1) time). What we've done is removed most of the cruft that surrounded the scheduling core (namely runqueues, 'jiffies', time slicing, etc). This instead uses a really awesome tree-balancing algorithm that should use less memory (hurray from embedded devices) and is "completely fair"; every process gets its "equal share" of time on the CPU.

GMorgan, on 10/11/2007, -1/+2Well it isn't surprising that the scheduler is still O(1). If it wasn't then the loss of performance and scalability would have offset any increased cleverness in the algorithm.

We always had an efficient algorithm. This one is just cleverer.

One thing it does is favour processes that give back chunks of their slices unused when they go to IO. Many schedulers punish IO bound tasks by not returning the CPU time they surrendered, this tries to rectify that. TritonX, on 10/11/2007, -0/+1Can you explain it with a car analogy?

tgoose, on 10/11/2007, -0/+7In all honesty, you don't need to at all. Just be happy in the knowledge that things might speed up in the future, without having to pay for new hardware!

GMorgan, on 10/11/2007, -0/+1It doesn't make the hardware better. The old algorithm was quite efficient as well. What it does is balance things more cleverly. Some things will be suited to other schedulers and I imagine they will have an in kernel option for picking the old one if you are in the class that prefers the old one.

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.