From owner-freebsd-current Sun Jan 28 15:07:05 1996 Return-Path: owner-current Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id PAA09313 for current-outgoing; Sun, 28 Jan 1996 15:07:05 -0800 (PST) Received: from phaeton.artisoft.com (phaeton.Artisoft.COM [198.17.250.211]) by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id PAA09308 for ; Sun, 28 Jan 1996 15:07:03 -0800 (PST) Received: (from terry@localhost) by phaeton.artisoft.com (8.6.11/8.6.9) id QAA01815; Sun, 28 Jan 1996 16:04:07 -0700 From: Terry Lambert Message-Id: <199601282304.QAA01815@phaeton.artisoft.com> Subject: Re: writes still slower than before recent vm changes To: bde@zeta.org.au (Bruce Evans) Date: Sun, 28 Jan 1996 16:04:06 -0700 (MST) Cc: current@FreeBSD.org In-Reply-To: <199601280713.SAA31441@godzilla.zeta.org.au> from "Bruce Evans" at Jan 28, 96 06:13:13 pm X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-current@FreeBSD.org Precedence: bulk > Bonnie reports that writes are 30%-50% slower than 2 weeks ago. However, > iozone reports that writes are slightly faster than 2 weeks ago. The > only significant difference between iozone writes and bonnie writes is > that iozone writes are page-aligned while bonnie writes aren't. Not > page aligning the buffers is stupid but shouldn't have such a large > penalty. Do either of them pre-create the file? If the file is precreated, a read must occur before a write can take place. The only "real" fix I can see for this would be to modify the cache to allow I/O in smaller than page-sized chunks, perhaps using a bitmap to indicate valid 512b chunks of a 4k page. This would at least reduce the amount of read required for a write. I don't know exactly what can be done about non-aligned boundries, though for a non alinged read of 512b or less, we are guaranteed 1/4 of the amount needed to read. This is still an improvement. I don't know how this would be impacted by a non-512b dev_bsize, nor how it should be altered to deal with the issue of frags. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.