[Gluster-users] poor performance of infiniband over tcp

lierihanmei lierihanmei at 163.com
Thu Nov 7 15:03:09 UTC 2013

I test performance of Infiniband rdma and tcp, using glusterfs 3.2.5. There are 10 servers and 2 clients, all of them been connected with Infiniband and have the same hardwares. Two hash volumes, each have 10 bricks,  All bricks are 16TB ext4 filesystem of raid5 on different servers. One volume's transport type is rdma and the other tcp. 

One client mounts the rdma volume and the other tcp volume. Using iozone to test the performance.
root at client-1:/mnt/rdma# iozone -i 0 -i 1 -s 10g -t 10 -R
root at client-2:/mnt/tcp# iozone -i 0 -i 1 -s 10g -t 10 -R

Performance of rdma volume:  
Read: 1.1GB/s 
Write: 1.0GB/s
While the performance tcp volume:
Read: 23.5MB/s 
Write: 52.8MB/s
I have checked the network and brick filesystem, all of them are normal. I wonder why performance of infiniband over tcp are so poor. 
Thanks for any help.

Best Regards,

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20131107/6e09e078/attachment.html>

More information about the Gluster-users mailing list