what I'm noticing is, with AFR/HA in gluster it seems that in the general case, 1.3 seems overall faster from a user experience perspective. 2.0 is a bit more robust and with that robustness seems to come a slight performance problem. when you do your benchmarking do so with some very large directories and some small ones, and some which change rapidly and some which change infrequently. it'll be interesting to see how that impacts things.. In my empirical observations, a large directory has a large impact in 2.0. it's entirely possible most of the performance related items I've notices are resolved, I'll need more time to see how things behave. At 07:17 AM 1/8/2009, Jerker Nyberg wrote: >Hi, > >The two interesting open source parallel distributed fault-tolerant file >systems I have found and tried are GlusterFS and Ceph. Perhaps Lustre will >be interesting for me in the future if they fix fault-tolerance without >special shared block storage on the servers. (Or have they?) Mirroring the >data two or more servers is good enough for me, although I would love some >RAID-6/RAID-Z -like redundancy. > >Ceph is found here <http://ceph.newdream.net/>. I did some benchmarking >for Ceph a few months ago <http://www.update.uu.se/~jerker/ceph/>. On Ceph >I got around 65 MByte/s write bandwidth on one node (using "dd") and >around 120 MByte/s aggregate for the whole cluster (7 nodes). > >It was around a year ago the last time I configured and ran GlusterFS on >the machines, but I do plan to do some better benchmarking for both file >systems in the not so distant future. > >--jerker > > >On Wed, 7 Jan 2009, a_pirania at poczta.onet.pl wrote: > > > I know that this is not the appropriate place :). You know someone can > > alternative to gluserfs ?:) > > > >_______________________________________________ >Gluster-users mailing list >Gluster-users at gluster.org >http://zresearch.com/cgi-bin/mailman/listinfo/gluster-users
