From owner-freebsd-stable Wed May 15 10: 4:39 2002 Delivered-To: freebsd-stable@freebsd.org Received: from damnhippie.dyndns.org (12-253-177-2.client.attbi.com [12.253.177.2]) by hub.freebsd.org (Postfix) with ESMTP id 1462437B407 for ; Wed, 15 May 2002 10:04:27 -0700 (PDT) Received: from [172.22.42.2] (peace.hippie.lan [172.22.42.2]) by damnhippie.dyndns.org (8.11.6/8.11.6) with ESMTP id g4FH4QG47451 for ; Wed, 15 May 2002 11:04:26 -0600 (MDT) (envelope-from freebsd@damnhippie.dyndns.org) User-Agent: Microsoft Outlook Express Macintosh Edition - 5.01 (1630) Date: Wed, 15 May 2002 11:04:28 -0600 Subject: Re: filesystem performance (was: enable/disable softupdates in rc init idea) From: Ian To: freebsd-stable Message-ID: In-Reply-To: <20020515204323.G19317-100000@woozle.rinet.ru> Mime-version: 1.0 Content-type: text/plain; charset="US-ASCII" Content-transfer-encoding: 7bit Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG > From: Dmitry Morozovsky >> But I've always been under the impression that softupdates helps with write >> performance, especially writing filesystem metadata, and very little writing >> happens on my root filesystems, I believe. > > A bit of. If you mount / with atime turned off. And if not -- each time > you access e.g. /etc/resolv.conf.... > > Sincerely, > D.Marck [DM5020, DM268-RIPE, DM3-RIPN] I use the noatime option on all my filesystems. If anyone knows of a good reason to not do that, feel free to let me know. (Other than the obvious reason, of course, that I can't see when a file was last accessed; I've decided I can live without ever knowing that.) Speaking of filesystem metadata performance, back in the FreeBSD 3.x days I discovered you got an order of magnitude type performance improvement by formatting a filesystem with fragsize == blksize; specifically, setting both to 8k. The reason for the improvement was that this eliminated a whole bunch of read-modify-write operations on directory and inode blocks. (It never quite made sense to me why the system wanted to RMW those blocks anyway, repeatedly issuing physical reads for the same blocks over and over when they should have already been buffered at the bio layer.) I'm curious whether this is still a factor in the 4.x world. I no longer have a conveniently instrumented custom scsi driver for monitoring physical sector-level IO like I had in 3.x, so I can't just poke it and see if it twitches anymore. -- Ian To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message