Date: Sun, 16 Dec 2001 22:36:52 +0100 From: Brad Knowles <brad.knowles@skynet.be> To: "Brandon D. Valentine" <bandix@looksharp.net>, Brad Knowles <brad.knowles@skynet.be> Cc: Terry Lambert <tlambert2@mindspring.com>, <hiten@uk.FreeBSD.org>, <chat@FreeBSD.org>, <grog@FreeBSD.org>, <phk@FreeBSD.org> Subject: Re: IBM suing (was: RMS Suing was [SUGGESTION] - JFS for FreeBSD) Message-ID: <a0510101db842c1faefc1@[10.0.1.22]> In-Reply-To: <20011214141902.F69086-100000@turtle.looksharp.net> References: <20011214141902.F69086-100000@turtle.looksharp.net>
next in thread | previous in thread | raw e-mail | index | archive | help
At 2:39 PM -0500 on 2001/12/14, Brandon D. Valentine wrote: > The major advantage people see to a journaling file system is that of > the lack of fsck on boot. This is crucial to large filesystems. But you've got this today with softupdates. Look at the filesystem work that Joe Greco has done for his multi-terabyte scale news servers, which he detailed at SANE 2000 (see <http://www.nntp.sol.net/sane2000>). He's got his file servers down to the point where he can flip the power switch off, flip it back on, and have the servers up and fully operational in less than thirty seconds. Moreover, the news system doesn't see a single user-visible hiccup, because it smoothly fails over to backup servers when the primaries become unavailable. You could probably do the same with AIX, but it would take a lot of work. > Perhaps when Softupdates on FFS gets to the point where snapshots and > background fsck are fully implemented and well tested and maybe even > enabled by default then people will stop asking for a journaling file > system. I don't know very much about JFS, but I do know that the design > of XFS offers some cool features like dynamic inode creation. XFS has a lot of cool features, but I don't see that we necessarily need to implement XFS per se, in order to get the same features. > One would > also think the ability to possibly relink rm'd files by rolling back > journal transactions would be a potentially useful feature on a > filesystem being used on the average user's desk. That only works up to the point where the journal rolls over, and assumes that all data writes as well as all meta-data writes are processed through the log. However, by forcing all data writes as well as meta-data writes through the journal, you increase by many orders of magnitude the amount of information that has to be written by the journal, and you also shorten its useful lifespan by the same amount. -- Brad Knowles, <brad.knowles@skynet.be> H4sICIFgXzsCA2RtYS1zaWcAPVHLbsMwDDvXX0H0kkvbfxiwVw8FCmzAzqqj1F4dy7CdBfn7 Kc6wmyGRFEnvvxiWQoCvqI7RSWTcfGXQNqCUAnfIU+AT8OZ/GCNjRVlH0bKpguJkxiITZqes MxwpSucyDJzXxQEUe/ihgXqJXUXwD9ajB6NHonLmNrUSK9nacHQnH097szO74xFXqtlbT3il wMsBz5cnfCR5cEmci0Rj9u/jqBbPeES1I4PeFBXPUIT1XDSOuutFXylzrQvGyboWstCoQZyP dxX4dLx0eauFe1x9puhoi0Ao1omEJo+BZ6XLVNaVpWiKekxN0VK2VMpmAy+Bk7ZV4SO+p1L/ uErNRS/qH2iFU+iNOtbcmVt9N16lfF7tLv9FXNj8AiyNcOi1AQAA To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-chat" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?a0510101db842c1faefc1>