Weekly Edition
Return to the Kernel page
| |||||||||||||
GlusterFS 2.0 released
The GlusterFS 2.0 release is
available. GlusterFS is an interesting cluster filesystem which runs
mostly in user space and which claims to scale into the petabyte range.
See the
feature list for an overview. "As your volume size grows beyond
32TBs, fsck (filesystem check) downtime becomes a huge problem. GlusterFS
has no fsck. It heals itself transparently with very little impact on
performance." License is GPLv3.
(Log in to post comments)
GlusterFS 2.0 released Posted May 17, 2009 20:25 UTC (Sun) by pkern (subscriber, #32883) [Link]
Reading the FAQ and documentation it looks to me that all scheduling decisions and replication are handled by the clients accessing the cluster. The servers seemingly serve as dumb bricks with no clue whatsoever what their peers are. Is that correct?
GlusterFS 2.0 released Posted May 18, 2009 5:43 UTC (Mon) by shehjart (guest, #58602) [Link]
Hi
Theoretically speaking, there is no concept of clients and servers in the GlusterFS design. However, we continue to use that terminology to stay closer to the traditional differentiation between nodes/machines when talking about distributed file systems.
The fact is, that there is no fixed definition of functionality that can be left only to the server and other functionality that can be left only up to the client. The various translators can be combined in different ways on both client and server machines to come up with various ways of distributing, replicating, unifying and striping your data, as per the needs of a specific deployment. So, "all scheduling decisions and replication", can also be handled at the server side if you so choose.
That said, in practice, we do recommend that certain translators be arranged with certain assumptions in mind in order to achieve maximum performance.
Finally, it is true that, by design, servers are not aware of what their peers are doing. This helps avoid any network communication between the bricks, helping us eliminate a major source of cluster scalability bottleneck, i.e. maintenance of consistent state among the servers/bricks. Of course, we could've gone down the meta-data server(MDS) approach, like Lustre and pNFS, but that too has limits in terms of resilience to MDS node failures.
Do come by #gluster on irc.gnu.org. We hang out there.
HTH.
GlusterFS 2.0 released Posted May 23, 2009 16:50 UTC (Sat) by kbob (guest, #1770) [Link] GlusterFS has no fsck.Pity. A generation of geeks will be deprived of the euphemism, Glusterfsck.
GlusterFS 2.0 released Posted May 26, 2009 17:21 UTC (Tue) by martinfick (subscriber, #4455) [Link]
"GlusterFS has no fsck."
That statement about glusterfs is a really strange one. Afterall, glusterfs is effectively a network filesystem and in most case defers to a sub filesystem such as EXT3 just like NFS. So, I would say this a misleading statement at best, and really has nothing to do with any of the strengths or weaknesses of this filesystem.
GlusterFS 2.0 released Posted May 27, 2009 5:49 UTC (Wed) by shehjart (guest, #58602) [Link]
A major motivation for putting it like that is the presence of replication functionality that also does automatic self-heal when a node goes down or there has been a network partition or even a disk crash on one of the nodes.
|
|||||||||||||