Date: Mon, 1 Oct 2001 19:57:17 -0500 From: Mike Meyer <mwm@mired.org> To: Carl Schmidt <carl@slackerbsd.org> Cc: questions@freebsd.org Subject: Re: accidently pulled the plug... Message-ID: <15289.4333.907347.872523@guru.mired.org> In-Reply-To: <93236649@toto.iv>
next in thread | previous in thread | raw e-mail | index | archive | help
Carl Schmidt <carl@slackerbsd.org> types: > On Mon, Oct 01, 2001 at 09:26:29PM +1000, Edwin Groothuis wrote: > > If you have softupdates enabled on your filesystems, then you will probab= > ly > > have the least problems. > Just a quick point regarding softupdates...softupdates is not the cure for > cancer many people like to make it out to be. It protects metadata and that= > is > all it protects. You are much more likely to lose data (NOTE: not metadata - > data) using softupdates if you are doing something disk intensive at the ti= > me > of the plug pulling or whatever other event causes a not-normal shutdown. Right. I was going to say something about that myself. There are actually *four* modes that you can mount file systems in. softupdates is the third safest. Safest first, it's: 1) sync. All data is written to disk synchronously. 2a) softupdates: Metadata is written synchronously, but ordered to increase performance and insure the integrity of the metadata. Data is still written asynchronously. 2b) noasync. Metadata is written synchronously, data asynchronously. This is the default mode. 3) async. Data and metadata are both written asynchronously. This is the default (and maybe only mode) for Linux. sync, noasync and async are flags at mount time. softupdates is a property of the file system. If softupdates is enabled, those flags are ignored. softupdates and noasync are both 2 because ordering the metadata writes takes a bit of time, increasing the danger of losing metadata. In doing so it makes sure the metadata on disk is in a consistent state, which means your disk is actually less likely to be damaged. Given those tradeoffs, some people consider softupdates more dangerous than noasync. I put it at 2a because it's nearly as fast as async. As a final note, if you are running a disk with write caching but not tagged queuing, the *drive* is running in async mode. It may be less safe, because Unix uses an algorithm that prevents a buffer kept indefinitely, but some drives don't. On the other hand, some drives make sure they have enough spare power around to guarantee the cache will be flushed if they lose power, so it's no problem. I have no idea how to tell which is which. <mike -- Mike Meyer <mwm@mired.org> http://www.mired.org/home/mwm/ Q: How do you make the gods laugh? A: Tell them your plans. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?15289.4333.907347.872523>