RDMA over the Internet Protocol Suite BOF (roi)

Monday, December 10 at 1530-1730
=================================

CHAIRS:	Allyn Romanow <allyn@cisco.com> 
	Stephen Bailey <steph@cs.uchicago.edu>

AGENDA:

1. Agenda Bashing                                   Allyn, Steph  5 min
2. Description of the problem being addressed
   domain, issues, justification, motivation        Steph         30 min
3. Discussion of various alternative approaches,
   including direct data placement                  Jeff Chase    30 min
4. Overview of current industry solutions,
   including RDMA                                   Jim Pinkerton 15 min
5. Requirements                                     Jim Wendt     20 min
6. Discussion of goals and proposed charter
   for a WG                                         Allyn         20 min



BOF Description:


Much of today's use of the Internet and IP networks is for
buffer-to-buffer data transfers, often in the form of bulk data or
transactional communications, using a variety of Internet protocols,
including NFS and CIS for file-oriented storage, and soon iSCSI for
block-oriented storage.


Gigabit and faster network communication using TCP/IP within data
centers can incur heavy system resource costs due to increased CPU and
system bus bandwidth utilization.  The copying of incoming data in
order to place it at its ultimate destination can dominate the
end-system overhead.


Direct data placement is a solution to this problem that has been
widely used on non-IP networks in research and industry.  Direct data
placement enables a network interface to place data from incoming
packets directly into buffers supplied by the target application. This
approach significantly reduces receive side overhead.


Until now the overhead of TCP/IP was important to a relatively small
community, primarily scientific supercomputing and database
applications.  Specialized non-IP solutions, such as VIA, have been
developed and have been shown to work well.  However, with the spread
of IP networks and the proliferation of data centers, using targeted
technologies for different high speed applications, including storage,
IPC, and normal IP data transfer, limits scalibility in the data
center.


A generic solution for low overhead high speed data transfer would
enable SCTP/ IP and TCP/IP networks to have lower overhead at a
reduced cost point while creating a single technology solution that
can be utilized for a variety of applications protocols.


There has been much research and industry experience with the related
issues and with various direct data placement protocols. This WG
proposes to investigate the issues and various solutions in order to
create requirements for a direct data placement protocol.


The proposed working group will develop two documents:
o A problem statement that describes the issues and why they are
   important. This document will also discuss some current industry
   solutions to the problem, including remote direct memory access
   technology (RDMA).  Informational RFC.
o A requirements document that describes the protocol requirements
   of a solution for low overhead, low latency, high bandwidth data
   transfer for use with IETF transport protocols. Informational RFC.


Reading:


http://ietf.org/internet-drafts/draft-garcia-direct-access-problem-00.txt
http://ietf.org/internet-drafts/draft-wendt-direct-access-reqs-00.txt


Mailing Lists:


General Discussion: rdma@yahoogroups.com
To Subscribe: rdma-subscribe@yahoogroups.com
Archive: http://groups.yahoo.com/group/rdma