Messages in this thread | | | Date | Mon, 28 Sep 1998 22:01:31 +0200 (CEST) | From | MOLNAR Ingo <> | Subject | Re: [PATCH] scheduler patch, faster still |
| |
On Mon, 28 Sep 1998 yodaiken@chelm.cs.nmt.edu wrote:
> Is it? And, isn't Richard arguing that his change makes sense only for > long queues?
aggreed. 'long runqueues' are definitely a red herring. Very big (Linux) servers and even artificial 'exploit' programs could not show any RL case where there are long runqueues. Eg. someone at Kiva (now they are Netscape) has asked the same questions and has run statistics on a very big webserver and the average runqueue length was around 3, maxing out at 15. (the article showing exact numbers can be found via DejaNews) Other reports show similar results. (i too tried to trigger very long runqueues for server type load, but failed). [the reason i tried to do some statistics was that i have a solution for the 'wake one' and 'thundering herd' theoretical problem, but under Linux it's a definit non-problem. I have tested an Apache test-setup which served some 1200 small static pages per second]
Long runqueues might be a problem on other OSs, but not on Linux. (or if yes, please show me the case)
(barring one case, when there are alot of CPU-bound processes around, in which case those processes waste so much cache resources anyway)
-- mingo
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |