I assume you're using the newly reworked AFR in the 1.4 branch?
The write-behind translator (on the client, at least) and the "o-direct
on" option of storage/posix seem to bring speeds back to something near
the old speeds. When I tried, they weren't quite stable, but I haven't
tried with the new TLAs from the last week (the last TLA I tried was when
the dht translator was just introduced). Last I heard, though, the new
AFR still needs a cleanup before it's considered stable.
Thanks,
Brent Nelson
Director of Computing
Dept. of Physics
University of Florida
On Sun, 28 Sep 2008, Brandon Lamb wrote:
I did see a post awhile back about the new AFR killing write speeds,
any update on this?
I setup a 2 server (both are server/client) and only mounted the afr
brick on one of them and attempted to untar linux-source-2.6.26.tar...
what took 5-9 seconds on the local disk took over 5 minutes on the afr
mount.
Are there any tricks I can employ? Is there an option to use such as
"i dont care about autonomous locks" or something? Hm I havent tried
WITHOUT afr, i'll go try that, but this seems kind of odd.
web2:/mnt/test# time tar xf linux-source-2.6.26.tar
real 0m6.393s
user 0m0.228s
sys 0m2.928s
web2:/mnt/test# time tar xf linux-source-2.6.26.tar
real 0m5.655s
user 0m0.220s
sys 0m3.836s
web2:/mnt/test# time tar xf linux-source-2.6.26.tar
real 0m9.639s
user 0m0.284s
sys 0m3.788s
web2:/mnt/test# time tar xf linux-source-2.6.26.tar
real 0m7.505s
user 0m0.272s
sys 0m3.792s
web2:/mnt/test# time tar xf linux-source-2.6.26.tar
real 0m6.865s
user 0m0.312s
sys 0m3.948s
web2:/mnt/test# cd ../gfs-import/
web2:/mnt/gfs-import# time tar xf linux-source-2.6.26.tar
real 5m26.856s
user 0m0.772s
sys 0m3.464s
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
