On 07/13/2012 05:08 PM, Rajesh Amaravathi wrote: > The issue was reported earlier, but I don't see any references it was > fixed in 3.3.0: > > > Deadlock happens when writing a file big enough to fill the > filesystem cache and kernel is trying to flush it to free some > memory for glusterfsd which needs memory to commit some > filesystem blocks to free some memory for glusterfsd... > > > http://gluster.org/pipermail/gluster-users/2011-January/006477.html > https://bugzilla.redhat.com/show_bug.cgi?id=GLUSTER-2320 > > **** > This is a problem generic to fuse/userspace filesystems. > Also, in 3.3, since we have NLM implemented to provide locking for NFS, > Its not possible to mount from a system which has glusterd(more precisely, > a Gluster NFS process) running since both kernel nfs and gNFS will try to register > for NLM v4 with portmapper. Hmm, this is sad. gluster/fuse mounts perform very poor for me. On a 10-server distributed gluster setup, when put to a lot of load, I'm able to serve (each server has 1xGbit NIC dedicated for gluster traffic and 1xGbit NIC dedicated for external traffic, webservers): - around 100 Mbit/s with gluster/fuse mounts to glusterd on localhost, - around 300 Mbit/s with NFS mounts to glusterd on localhost, - around 500 Mbit/s with NFS mounts to glusterd on localhost, and fsc/cachefilesd. Killing the option to use NFS mounts on localhost is certainly quite the opposite to my performance needs! -- Tomasz Chmielewski http://www.ptraveler.com
