RE: Architectural scope of IP over Inifiniband

Ok, first off, IPoIB is being defined over Unreliable Datagram.  So it
is not taking advantage of the IB reliable protocols - and not using
those resources either.

The intent is that many applications are written to the BSD Sockets
model, so allowing these applications to seamlessly use IB as a layer 2
is important.  

I guess I would ask you what is your intent - to tunnel packets in
Ethernet until you get somewhere outside that supports TCP/IP ?  The
answer is NO, you expect IP to be supported everywhere, and to be able
to layer UDP, SCTP, and TCP on top of those semantics.

That said you can look at IB as a VERY overengineered layer 2 (and that
is my personal opinion for all things IPoIB) or as a layer 4 networking
stack that you can write native applications for (See RDDP vs. IB/RDMA)

Bill (spoken with the WG chair hat firmly off)

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf
Of [email protected]
Sent: Saturday, September 13, 2003 3:10 PM
To: [email protected]
Subject: [Ipoverib] Architectural scope of IP over Inifiniband




I was actively following the RDDP group, and happened
to come across this interesting group as well.

I am kind of lost as to why one would want IP which is unreliable to be
carried over a reliable link infrastructure without any reliable
transport layer at the TCA or the HCA.

Would the intent of this be to tunnel the IP packets
through a Infiniband fabric and emerge as IP to some
point outside that supports TCP/IP for reliable
transport.

------- End of forwarded message -------

_______________________________________________
IPoverIB mailing list
[email protected] https://www1.ietf.org/mailman/listinfo/ipoverib