From owner-freebsd-stable Sat May 5 17:45:26 2001 Delivered-To: freebsd-stable@freebsd.org Received: from vimfuego.saarinen.org (saarinen.org [203.79.82.14]) by hub.freebsd.org (Postfix) with ESMTP id 0F58737B422 for ; Sat, 5 May 2001 17:45:23 -0700 (PDT) (envelope-from juha@saarinen.org) Received: from dendennis.saarinen.org ([192.168.1.2] helo=dendennis) by vimfuego.saarinen.org with smtp (Exim 3.22 #1 (Red Hack)) id 14wC59-0000k7-00; Sun, 06 May 2001 12:07:27 +1200 From: "Juha Saarinen" To: "Doug Russell" , "Matt Dillon" Cc: "freebsd-stable" Subject: RE: soft update should be default Date: Sun, 6 May 2001 12:06:55 +1200 Message-ID: MIME-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit X-Priority: 3 (Normal) X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook IMO, Build 9.0.2416 (9.0.2911.0) In-Reply-To: Importance: Normal X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4133.2400 Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG :: Yes, yes, yes! No argument there. People should realize what they are :: doing by enabling the write cache, softupdates or no softupdates. In my experience, very few people even know that there is such a thing as a write-cache setting on hard drives. Also, most hard drives today ship with write-caching enabled, without tools to turn it off. Someone mentioned a sysctl that can be used to turn off drive write-caching... which one is it? I asked my friend Daryl Jay, who as the tech manager of a systems integrator has had plenty of experience with hard drives in various situations, about write caching. Daryl says WCE has always been a problem; Windows 9x and ME can suffer file system corruption with IDE hard drives that have large caches, as they can become corrupted during an APM/ACPI shutdown -- MS issued a patch to for Win98SE and WinME, but not fo the older variants, that increases the delay before power-off to permit the cache contents to be written out. He also confirms that if you have a SCSI RAID controller with onboard cache, and set WCE for the drives / array as a whole, you have to have a UPS or a battery to avoid f/s corruption. However, a transaction / journalled file system such as NTFS, XFS and ReiserFS goes a long way towards mitigating the effects of incomplete cache write-outs. OTOH, FAT16/32 and Linux Ext2fs are very vulnerable in this respect. No idea how UFS w/ or w/o softupdates performs here, unfortunately. Daryl says that IBM drives are less likely to cause corruption, as they shipped with a decently-sized power capacitor which in most cases will have enough power to complete the cache write-out. This doesn't help with caching SCSI controllers, obviously. IBM used to ship all its drives with WCE set to off, and charged a fee to have the drive "AV optimised", which was just a simple procedure to turn WCE on. As of 2-3 years ago, all IBM drives are shipped with WCE on, however. The performance drop you see when you turn off WCE is substantial. Daryl and a customer tested a 36GB IBM Scorpion with and without WCE... without, they got 3.2MBps average sustained DTR across the disk, with, 17.8MBps. Just a few datapoints, in case anyone's interested... I think this should be documented in the manual. -- Juha To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message