From owner-freebsd-current@FreeBSD.ORG Wed Aug 31 01:55:14 2005 Return-Path: X-Original-To: freebsd-current@FreeBSD.org Delivered-To: freebsd-current@FreeBSD.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id C2FE116A41F for ; Wed, 31 Aug 2005 01:55:14 +0000 (GMT) (envelope-from truckman@FreeBSD.org) Received: from gw.catspoiler.org (217-ip-163.nccn.net [209.79.217.163]) by mx1.FreeBSD.org (Postfix) with ESMTP id 7AC8143D46 for ; Wed, 31 Aug 2005 01:55:14 +0000 (GMT) (envelope-from truckman@FreeBSD.org) Received: from FreeBSD.org (mousie.catspoiler.org [192.168.101.2]) by gw.catspoiler.org (8.13.3/8.13.3) with ESMTP id j7V1t57O016802; Tue, 30 Aug 2005 18:55:09 -0700 (PDT) (envelope-from truckman@FreeBSD.org) Message-Id: <200508310155.j7V1t57O016802@gw.catspoiler.org> Date: Tue, 30 Aug 2005 18:55:05 -0700 (PDT) From: Don Lewis To: doconnor@gsoft.com.au In-Reply-To: <200508311045.56657.doconnor@gsoft.com.au> MIME-Version: 1.0 Content-Type: TEXT/plain; charset=us-ascii Cc: freebsd-current@FreeBSD.org, kabaev@gmail.com Subject: Re: Odd performance problem (hitching) X-BeenThere: freebsd-current@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Discussions about the use of FreeBSD-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 31 Aug 2005 01:55:14 -0000 On 31 Aug, Daniel O'Connor wrote: > On Tuesday 30 August 2005 17:50, Daniel O'Connor wrote: >> > You might try significantly decreasing MAXVNODES_MAX in >> > sys/kern/vfs_subr.c and rebuilding your kernel. >> >> OK.. >> I'll try timing when the drop outs happen :) > > Hmm they appear to be every 10 seconds on the dot. Hmn, I wonder what runs every 10 seconds ... > I haven't changed any syncer delay stuff either.. > kern.filedelay: 30 > kern.dirdelay: 29 > kern.metadelay: 28 Those won't make any difference. They govern how many slots behind in the worklist that items of various types get inserted. The syncer wakes up once per second, processes all the items in a worklist slot, then goes back to sleep. The syncer vnode for each file system gets inserted when the file system is mounted and stays in the same slot, so the ugly FOREACH loop I mentioned earlier gets executed once each time the entire worklist is scanned, and since the worklist has 32 slots, the problem should only occur once every 32 seconds if you have one file system that has hogged most of the vnodes. If there are multiple file systems, the syncer vnodes will inserted at various power of two locations in the list, but I don't see a way to get activity spaced at ten second intervals. > Still using SCHED_4BSD too. >