I tried to parse your English, as it seems not to be your first language, but I've grokked that you are basically having issues with NFS performance. One thing I've tried is to run multiple NFS daemons per server, this can easily be done with CTDB, multiple IP addresses per server, and one NFS daemon for each IP address. The hard part of this is to override the default NFS daemon started by glusterd and manage the daemons outside of glusterd. I've used runit for this, but any decent init system should work. I've seen this scale NFS performance significantly and enable much better CPU utilization in a high IOP environment. Besides this work around, I would also be interested in the progress on epool. Have NFS daemons limited to processing events on a single thread is truly a bottleneck in high IOP environments. Thanks and best of luck! Matthew McKeen matthew@xxxxxxxxxxx On Sat, Apr 11, 2015 at 8:35 PM, 张兵 <cust004114@xxxxxxx> wrote: > Hi all, > How To enhance the gluster performance of NFS, when EC, the network > transmission time long processing time, effect of NFS IOPS; > In our tests, each EC IO needs 1ms(Gigabit Network), NFS can only > handle 1000 IOPs ;NFS protocol message and the process in the same > process,The process will block the NFS protocol message received; > Why not the received nfs message packet and nfs protocol processing > process isolation Or NFS multi thread; > By the way, the progress of gluster epool multithread, if use epool > multithread ,can solve the problem gluster nfs performance.? > Best regards > > > 您是不是要找:cluster nfs > > > > > > _______________________________________________ > Gluster-users mailing list > Gluster-users@xxxxxxxxxxx > http://www.gluster.org/mailman/listinfo/gluster-users -- Matthew McKeen matthew@xxxxxxxxxxx _______________________________________________ Gluster-users mailing list Gluster-users@xxxxxxxxxxx http://www.gluster.org/mailman/listinfo/gluster-users
