[Gluster-devel] Gluster application
greno at verizon.net
Tue Feb 27 13:21:17 UTC 2007
This is my initial attempt at clustering outside of just 2-node HA.
Originally I think that OCFS2 was targeted as providing access to shared
storage array such as iSCSI but I think it is becoming more general.
GFS appears somewhat more advanced right now than OCFS2 but is more
complex. But again I have not installed either of these, just from my
readings. The goal for me is to have all files/data be local and not
just shared access so that there are no io bottlenecks during access and
io performance will be very high. For this GlusterFS seems like it
might work. My cluster filesystem has to be able to ride on top of LVM2
and I would like that this is simplified by an LVM2 management layer so
that I don't have to change the configs on every single cluster node
whenever I want to expand/contract a logical volume. I should be able
to change the LVM2 config on just one node and it should automatically
migrate to all other nodes dynamically. This is missing with
GlusterFS. I think I can have this with GFS with their lvm2-cluster
management tool. OCFS2 appears able to migrate the LVM2 metadata but
only during cluster startup so you have to bring the cluster down except
for one node to make any changes to LVM2. You make the LVM2 changes on
the one node and then you can restart the cluster and those changes
should be migrated throughout the cluster. This is what my research
shows anyway. So since GlusterFS is not tightly tied into the kernel
perhaps they will not be able to do such things. But maybe there can be
other tools that could be used to fill in the gap. Something like a
generic lvm2-cluster tool that would work for other types of clusters
not just GFS. I'm going to wait for Feb 28 release before I start
experimenting with GlusterFS. I'll probably try to setup some testbed
using Xen doms.
jeffrey macdonald wrote:
> I use FC6 as well, but not the xen kernel. It works
> but i was not able to get the performance/io-threads stuff working (yet).
> I was going to try again with the next pre-release.
> fuse loads a kernel module but nothing else does.
> Please correct me if im wrong but as i understand it.
> GFS and OCFS2 are for file systems accessing
> a shared storage array. glusterfs is more for machines
> that have their own direct attached storage, but no shared
> storage array.
> Please let me know how your FC6 experience with
> glusterfs is.
More information about the Gluster-devel