From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S261977AbTHYRMS (ORCPT ); Mon, 25 Aug 2003 13:12:18 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262112AbTHYRMS (ORCPT ); Mon, 25 Aug 2003 13:12:18 -0400 Received: from 153.Red-213-4-13.pooles.rima-tde.net ([213.4.13.153]:18949 "EHLO small.felipe-alfaro.com") by vger.kernel.org with ESMTP id S261977AbTHYRMM (ORCPT ); Mon, 25 Aug 2003 13:12:12 -0400 Subject: Re: [RFC] renicing X From: Felipe Alfaro Solana To: Nick Piggin Cc: linux-kernel In-Reply-To: <3F4A3293.8070004@cyberone.com.au> References: <3F4A3293.8070004@cyberone.com.au> Content-Type: text/plain Message-Id: <1061831528.2967.7.camel@teapot.felipe-alfaro.com> Mime-Version: 1.0 X-Mailer: Ximian Evolution 1.4.4 Date: Mon, 25 Aug 2003 19:12:08 +0200 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 2003-08-25 at 18:00, Nick Piggin wrote: > My scheduler patch really benefits a lot from renicing X. I > think its because it nices more nicely. Any reasons why this > might be a bad idea? Well, not for me... Although renicing X with Con patches makes X feel horrible, with your patches is not as horrible. However, I feel X much smoother with X reniced at +0. Renicing X at -20, for example, may reduce mouse cursor jumpiness under load, but makes X feel a little jerky in general (window movement is not as smooth as with X niced at +0). This is, however, based on subjective testing, not actual numbers. But for interactivity, most of the time it's the subjective feeling of the user about the system what matters, not numbers. And now we're talking about sched-policy-7a: under heavy load, spawning new processes still takes twice the time it takes when the system is under no load. For example, spawning a new Konsole session (not a new Konsole process, but a new session tab inside Konsole) takes approx. 1 second on my P3-700Mhz. However, with sched-policy-7a and under heavy load (the mad while true; do a=2; done loop), it takes more than 2 seconds. In general, sched-policy-7a feels extremely smooth and responsive in general but, for me, Con patches offer the smoothest X experience I have ever felt until date. Anyways, I will keep testing your patches and I greatly encourage you to keep improving them. It's always good to have diversity :-)