From owner-freebsd-arch Mon Feb 10 19:14:42 2003 Delivered-To: freebsd-arch@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id E5B5337B401 for ; Mon, 10 Feb 2003 19:14:39 -0800 (PST) Received: from web13404.mail.yahoo.com (web13404.mail.yahoo.com [216.136.175.62]) by mx1.FreeBSD.org (Postfix) with SMTP id 2BEDC43FBD for ; Mon, 10 Feb 2003 19:14:39 -0800 (PST) (envelope-from giffunip@yahoo.com) Message-ID: <20030211031438.42114.qmail@web13404.mail.yahoo.com> Received: from [200.24.79.141] by web13404.mail.yahoo.com via HTTP; Tue, 11 Feb 2003 04:14:38 CET Date: Tue, 11 Feb 2003 04:14:38 +0100 (CET) From: "=?iso-8859-1?q?Pedro=20F.=20Giffuni?=" Subject: FFS performance improvements? To: freebsd-fs@FreeBSD.org MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit Sender: owner-freebsd-arch@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG (sorry for the crossposting, replies should probably go to -fs) Hi; I heard a complaint on the FreeBSD lists about journalling being faster than softupdates. While I'm not extremely interested in comparisons with other fs's that I can't use, I do recall it is possible to enhance FFS speed with the tricks used in CFFS. JIC ppl forgot, and since it is a good reference, the classic document is here: http://www.pdos.lcs.mit.edu/pubs.html (under Storage Management) Of the two tricks explained there, explicit grouping gave the most significant performance improvements, without causing problematic side effects. I've been "bugging" the only two persons I know that might have some patches for this with the following results: __________ I considered doing the co-location game in UFS2, but in the end decided that the performance gain was not sufficient to take the effort to do it. So, I put my effort into other things. Kirk McKusick __________ C-FFS required changes to OpenBSD so that the buffer cache would cache raw disk blocks and would consult those cached raw disk blocks before going to the disk device. I do have some old patches and I'll try to dig them up. It was my first kernel project for BSD unix and is definitely nowhere close to production quality. (probaby lots of mistakes related to concurrency and resource management) http://pdos.lcs.mit.edu/~csapuntz/cffs.tar.gz -Costa __________ So, my questions are... 1) Although porting C-FFS is probably not the way to go, do we need to modify the VM to fulfill the requirements, or perhaps this is something GEOM can do? (sorry I don't have clear exactly what GEOM does, but it would seem somewhat related to caching raw disks) 2) Since I'm not really into kernel hacking, if I do find the patches for FFS is someone interested in playing with them in exchange for improvements of "just" between 10-50% with small files ?? ;). cheers, Pedro. ______________________________________________________________________ Yahoo! Cellulari: loghi, suonerie, picture message per il tuo telefonino http://it.yahoo.com/mail_it/foot/?http://it.mobile.yahoo.com/index2002.html To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-arch" in the body of the message