Subject: Re: Revising CPU usage calculations
To: SODA Noriyuki <soda@sra.co.jp>
From: Jason Thorpe <thorpej@shagadelic.org>
List: tech-kern
Date: 03/28/2007 10:26:28
On Mar 28, 2007, at 4:48 AM, SODA Noriyuki wrote:
>>>>>> On Wed, 28 Mar 2007 13:04:28 +0200,
> <dsieger@techfak.uni-bielefeld.de> said:
>
>> ok, gathering from what has been written so far:
>>
>> 1. Can we agree that autonice should go into SCHED_4BSD only?
>
> I don't think so.
> Is there any reason that autonice is SCHED_4BSD only?
> The idea itself (i.e. automatically add a penalty to a CPU hog)
The irony here is that the 4BSD scheduling algorithm is ALREADY
supposed to penalize CPU hogs by decaying their priority based on
their CPU usage.
> can apply any timesharing schedulers, and 4BSD scheduler doesn't
> have to have the autonice feature at all. There are many systems
> which are using 4BSD scheduler but doesn't have autonice.
> For example, FreeBSD removed autonice more than 10 years ago
> (kern_synch.c revision 1.25).
Ah, another good reason to get it out of our scheduler! :-)
>
>> 2. As there seem to be some good arguments for not using autonice at
>> all, what about making it a sysctl tunable for the 4BSD scheduler?
>
> Well, it's already sysctl tunable, and default off, isn't it?
> --
> soda
-- thorpej