On Wed, 16 Feb 2005, Norman Zhang wrote: > I'm not sure how's XFS compared to other fs. I've been using it for few > years. It seems to be fine.
It works fast as all heck, but your data will *NOT* necessarily survive a crash (zeros everywhere if the data was not on disk). The filesystem metadata will survive the crash just fine, so you might not even know where the corruption is. So do not use it where you can reasonably expect a crash to happen, if the data in it ain't transient. IMHO, anyway. That does mean do not run it in a machine without SECDED memory or Chipkill memory (i.e. run it only where you have working ECC single-bit correction in hardware, or something better). Every so often (like once an year) I'd run a xfs_repair on it, just in case a kernel bug or memory corruption is silently screwing things up (if it gets bad enough for XFS to notice it, it will *dead lock* that filesystem, and hang everything that touches it). It has happened to me in the past, but that was more than one year ago, in Linux, and XFS has come a long way since then. All that said, I quite like XFS, and I use it a lot *except* on the root filesystem (which is always ext3). So far, nothing that uses fsync() properly ever lost a single byte, and that covers the MTAs and Cyrus. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
