[Gluster-users] Delayed Replication + Realtime Locking
vikas at gluster.com
Mon Jul 20 11:03:54 UTC 2009
----- "Mathew Eis" <mathew at eisbox.net> wrote:
> 1) Is the above possible with the current Gluster translators? I've
> experimented with various combinations of translators, including
> replicate, nufa, writebehind, and io-threads, and while I have been
> able to improve performance of some operations, many others are still
> taking place at WAN speed.
This should be possible with the replicate translator. Replicate returns
successfully to the application as soon as a write succeeds on atleast
one subvolume. So, in effect, the write over WAN should happen in
the background. You should also set the read-subvolume to be the local
server, so that reads are served from there.
However, currently there is one feature that makes a setup like this
somewhat untenable. When a file is self healing, all access to it
will be blocked. Self-heal over WAN could turn out to be quite slow
for large files. There are two improvements planned for the 2.1 release
which will make your configuration possible:
1) Self heal in the background. With this only writes to the currently
self-healing region of a file will be blocked.
2) rsync algorithm for self-heal. This will be much more efficient in terms
of data transferred over the network.
More information about the Gluster-users