Date: Tue, 07 Mar 2006 20:46:30 +0100 From: Ivan Voras <ivoras@fer.hr> Cc: freebsd-hackers@freebsd.org Subject: Re: NetBSD disk backup over network Message-ID: <440DE316.8060604@fer.hr> In-Reply-To: <20060307162014.GO17589@over-yonder.net> References: <25525887.1141742017684.JavaMail.root@vms172.mailsrvcs.net> <20060307162014.GO17589@over-yonder.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Matthew D. Fuller wrote: >>You can easily save the stream of updates as a redo log (well, >>that's the idea I've been running around with). > > Isn't that what the gjournal SoC thing was about? No, not exactly. The idea was to make a journal of a GEOM device I/O requests on a separate device in the attempt to solve filesystem journaling, but that didn't work out (actually, gjournal works more or less fine, but as far as I understand it, there's a problem similar as with SoftUpdates - even if you journal writes on device level with filesystem mounted "sync", UFS keeps references to inodes in not-entirely-consistent way, so a fsck is always needed after unclean shutdown; in other words, UFS journaling must be done on UFS/VFS level, not GEOM in order to keep track of metadata semantics). Something like a filesystem that doesn't do any read caching, and keeps ALL data ALWAYS in sync with on-disk state could work with gjournal and also with gmirror+geom_gate combination. Such filesystem would probably be similar to GFS, AFAIK.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?440DE316.8060604>