[ovs-discuss] Q: RDMA tunnel support in openvswitch

[Alexey I. Froloff]

On Tue, Oct 02, 2012 at 11:16:31AM -0700, Jesse Gross wrote:
> Where specifically do you think that this will improve performance?
> I don't know Infiniband well so it's possible that I'm wrong
> but I suspect that you will end up with something very similar to the
> existing network infrastructure.
IPoIB (and esp. EoIB) have significant overhead.  On 56 Gb/sec
(4X FDR) link we can't have more than 10-12 Gbps between two
hypervisors using STT tunnels via IPoIB interfaces.  And high sys
load, because IPoIB incapsulation performed in kernel.

Passing memory directly from hypervisor to hypervisor should
greatly improve performance.

> The other issue that I see is that presumably this involves some form
> of protocol to transmit the encapsulated Ethernet packets.  Is there
> an existing protocol that you were planning on using?
What kind of information (besides the packet itself) must be
transferred to remote node?  I am only thinking about "tunnel
id" (for GRE in_key/out_key is used).  So encapsulation is
pretty simple.

-- 
Regards,    --
Sir Raorn.   --- http://thousandsofhate.blogspot.com/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20121002/2e02215d/attachment.sig>