From owner-freebsd-questions Fri Jul 24 03:04:19 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id DAA04066 for freebsd-questions-outgoing; Fri, 24 Jul 1998 03:04:19 -0700 (PDT) (envelope-from owner-freebsd-questions@FreeBSD.ORG) Received: from boco.fee.vutbr.cz (boco.fee.vutbr.cz [147.229.9.11]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id DAA04052 for ; Fri, 24 Jul 1998 03:04:09 -0700 (PDT) (envelope-from cejkar@dcse.fee.vutbr.cz) Received: from kazi.dcse.fee.vutbr.cz (kazi.dcse.fee.vutbr.cz [147.229.9.51]) by boco.fee.vutbr.cz (8.9.1/8.9.1) with ESMTP id MAA14754 for ; Fri, 24 Jul 1998 12:03:35 +0200 (MET DST) Received: (from cejkar@localhost) by kazi.dcse.fee.vutbr.cz (8.8.8/8.8.8) id MAA08330 for freebsd-questions@freebsd.org; Fri, 24 Jul 1998 12:03:34 +0200 (CEST) From: Cejka Rudolf Message-Id: <199807241003.MAA08330@kazi.dcse.fee.vutbr.cz> Subject: Re: Bug in the process-scheduler & niceness of 20? To: freebsd-questions@FreeBSD.ORG (freebsd-questions@freebsd.org) Date: Fri, 24 Jul 1998 12:03:34 +0200 (CEST) X-Mailer: ELM [version 2.4ME+ PL31 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-questions@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG > On Fri, 17 Jul 1998, Cejka Rudolf wrote: > > > Yes, ok. But why is niceness control _so very_ insensitive? > > How can I increase nice sensitivity? > > Hack the kernel? Perhaps - I don't know anything else. Is this easy (one place, number change) or more elaborate change? > > Example: > > > > 1. process with nice = 0: CPU = 65 % > > 2. process with nice = 20: CPU = 30 % > > > > Ratio 2:1 is _too_ small. How can I reach ratio 5:1 or bigger? > > idprio is unusable for this purpose. Time ratio reacheble with idprio > > is 1:1 or infinity:none. > What are you running that requires so much precision? "Precision" isn't good word, "ratio" is right. Command idprio is in most cases unusable because the regular user isn't allowed to use this command. So only nice remains. If I run computing process via idprio yet (rc5des as example) and another computing process with normal priority (ray-tracing), idprio process stops. It's good and I like this. But there are some cases in which I want to donate about 5 % (or 10 %) of processor time constantly for any low priority process (not only rc5des exists). Just because to know something will be computed after a while (with bouded time) :-) With nice I have made some tests for two main running (computing) processes: 1: nice = 0 2: nice = 19 FreeBSD => 2:1 64 % 30 % Sun => 3:1 74 % 24 % Linux => 10:1 86 % 9 % Yes, I like many featuers of FreeBSD task scheduler - especially perfect response time. But 30 % for niced process is too much for me. My idea follows: Processor usage for two processes (after one minute) NICENESS "./proc & nice -NICENESS ./proc &" 0 48 % 48 % (~ 1:1) 5 59 % 37 % (~ 2:1) 10 70 % 26 % (~ 3:1) 15 81 % 15 % (~ 6:1) 20 91 % 5 % (~ 20:1) Are there any counterparts against this idea? -------------------------------------------------------------------------- Rudolf Cejka (cejkar@dcse.fee.vutbr.cz; http://www.fee.vutbr.cz/~cejkar) Technical University of Brno, Faculty of El. Engineering and Comp. Science Bozetechova 2, 612 66 Brno, Czech Republic To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message