By nature of how fuse/VFS works, each file involves a seperate network call, which is a lot of penalty compared to disk based filesystems. avati 2008/1/4, Sascha Ottolski <ottolski@xxxxxx>: > > Am Freitag 04 Januar 2008 17:08:00 schrieben Sie: > > what about the second run of du -hs via glusterfs? > > okay, I did several "time du -sh /mnt/gluster/somedir" > > second run is faster by 30 to 60 seconds. but then, it still takes about > 30 > seconds on one client to get the result, and over 2.5 minutes on the > other. > the latter one is a client that is in a long running process copying many > small files to the gluster. the first is idle. > > still, if i do the same directly on the servers, it takes only 0.02seconds. > > > Thanks, Sascha > > > > > > avati > > > > 2008/1/4, Sascha Ottolski <ottolski@xxxxxx>: > > > Hi, > > > > > > I'm wondering, has this to do with the disabled stat-prefetch > translator? > > > I > > > have a unify over 2 afr (2 bricks each) volume, spanning 2 servers > that > > > afr > > > each other. A "du -sh somedir" takes no time on each server directly > (one > > > has > > > 82 MB, the other 84 MB in somedir, coming from ~10.500 files files on > > > each server (in a heavy neseted directory structure). > > > > > > If I do a "du -sh subdir" on the client, it takes like some minutes > until > > > the > > > 166 MB result is ready. > > > > > > Test is done on latest tla checkout. > > > > > > > > > Thanks, Sascha > > > > > > > > > _______________________________________________ > > > Gluster-devel mailing list > > > Gluster-devel@xxxxxxxxxx > > > http://lists.nongnu.org/mailman/listinfo/gluster-devel > > > -- If I traveled to the end of the rainbow As Dame Fortune did intend, Murphy would be there to tell me The pot's at the other end.
