Many file servers using disk drives are "unreliable" if you shut them down without saving data in memory first. Both /sys/src/fs and disk/kfs force certain metadata updates to disk first to try to ensure that the fs structure, if not the content, remains consistent. In fact, disk/kfs does more than the old file server (it forces indirect block updates out as well). Disk/kfs only seemed less reliable than the old file system because it is more likely to be shut down without syncing. Also, the old file server could be recovered from a consistent root if the worm dump was used. If you sync, you shouldn't have too much trouble. Of course, forcing write-through makes the file system slower than some others for updates. At the cost of some code complexity, fossil attempts to do better, using a "soft update" scheme to ensure or attempt to ensure that everything on disk is consistent. It's mentioned briefly in the paper.
On 25 December 2011 16:30, Aram Hăvărneanu <ara...@mgk.ro> wrote: > I > have heard very good things about the reliability of both. You say > kfs is not very reliable, oh well. >
