Talk:Completely Fair Scheduler

O(1) vs. O(log n) performance
I guess the O(log n) CFS was introduced to the Linux kernel for performance improvements compared to the older O(1) scheduler, the computational resource compared being number of clock cycles/time. Yet, there should be a minimum number of processes (n) for which the performance of the O(log n) scheduler starts getting worse compared to a rather high-overhead, but still constant scheduler. Maybe this is for currently pathological examples, like thousands of processes. Can anyone say a few words about this? Thanks, --Abdull (talk) 08:20, 7 October 2010 (UTC)


 * Actually the main reason for the new scheduler was not "performance" (e.g. scheduler overhead), but fairness. When CFS was first merged to mainline, it was actually slower than the O(1) scheduler even at low loads. However, compared to everything else that needs doing, scheduler overhead is usually insignificant. CFS has also been optimized since then, so it's possible that it's faster than the old scheduler now. But I don't know of any precise measurements of this. -- intgr [talk] 15:06, 7 October 2010 (UTC)

Reverting BFS edit ('RV wrong eid)')
Sorry for the terrible commit message, accidentally pressed enter while still typing it. Meant to say: CFS was not inspired by BFS, nor could it have been -- BFS was begun over 2 years after CFS (August 2009 vs April 2007). As Molnár's email says, it was inspired by Kolivas's Staircase Deadline scheduler. See https://lwn.net/Articles/230574/. -- simxp (talk) 14:38, 15 April 2016 (UTC)

Revert controversy section
The controversy section is actually very useful to understand why this article is relevant anyway. I see that it was removed by this change:

(cur | prev) 08:33, 10 May 2015‎ 96.57.23.82 (talk)‎. . (10,386 bytes) (-1,771)‎. . (The section serves no purpose to understanding CFS. Maybe it should be in someones bio pages.) (undo) (Tag: section blanking)

I would dare to say that CFS has "relevance" because of the controversy between the Linux developers, a discussion that Linus Torvalds seems to have lost (see https://www.quora.com/Has-anyone-ever-won-an-argument-with-Linus-Torvalds).