[Gluster-devel] IB/RDMA Testing for 3.4
rgowdapp at redhat.com
Fri Dec 28 07:34:49 UTC 2012
----- Original Message -----
> From: "Andrew Holway" <a.holway at syseleven.de>
> To: "Raghavendra Gowdappa" <rgowdapp at redhat.com>
> Cc: gluster-devel at nongnu.org, "Иван Димитров" <dobber at amln.net>
> Sent: Thursday, December 27, 2012 3:28:58 AM
> Subject: Re: [Gluster-devel] IB/RDMA Testing for 3.4
> >> Is the Gluster RDMA stuff derived from the NFS over RDMA stuff?
> > Yes. Largely implemented based on RFCs 5666 and 5667.
> Does gluster use RDMA for client connections or internal stuff or
> both? Googling didn't come up with much.
As of now, tcp/ip is used for connection establishment b/w client and (glusterd, bricks). Once connection is established, all data goes through infiniband network. Effectively all the file system data goes through infiniband. Message sizes exceeding certain threshold are transferred using rdma and others through send/recv.
> > regards,
> > Raghavendra.
More information about the Gluster-devel