The returns for the effort of increasing the fuse channel size is hardly anything. Though you will see a nice increase in performance when block size is increased, because of the reduction of context switches, there will be a peak after which increasing the block size further has no benefits. Since a transaction involves memcpy of the buffers, the larger block sizes start becoming less friendly with your hardware memory caches and start working against your performance. >From our tests 128KB (the default maximum) is very ideal.
Avati On Sun, Oct 11, 2009 at 6:55 AM, <mki-gluste...@mozone.net> wrote: > Hi > > Are there any caveats to changing the max_read fuse mount option that's > hardcoded into xlators/mount/fuse/src/fuse-bridge.c to something other > than 128k? > > On my client test system, cat /proc/mounts shows: > > fusectl /sys/fs/fuse/connections fusectl rw,relatime 0 0 > 10.10.10.11 /storage fuse.glusterfs > rw,relatime,user_id=0,group_id=0,default_permissions,allow_other,max_read=131072 > 0 0 > > This is a deb (lenny) based x86_64 system, but the kernel is vanilla > 2.6.31.1 although the library is the default apt based > libfuse-dev_2.7.4-1.1 one. > > Thanks. > > Mohan > _______________________________________________ > Gluster-users mailing list > Gluster-users@gluster.org > http://gluster.org/cgi-bin/mailman/listinfo/gluster-users > _______________________________________________ Gluster-users mailing list Gluster-users@gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users