[dtn-interest] notes from IETF discussion on the future of the DTN reference implementation

[Christopher Small]

I gave a short talk at the DTNRG meeting at IETF on Thursday March 13, 
pointing out that the future development and maintenance of the DTN 
reference implementation (DTN2) was uncertain, and that we, as a 
community, should discuss what we would like from a reference 
implementation, and who is going to develop and maintain it.

Slides from my talk are archived at 
http://www.ietf.org/proceedings/08mar/slides/DTNRG-3.pdf.

After the meeting, seventeen of us convened to discuss the topic. 
Attending were

Christopher Small, BBN Technologies
Scott Burleigh, JPL
Keith Scott, MITRE
Mike Petkevich, Lockheed Martin
Hans Kruse, Ohio University
Jim Wyllie, Ohio University
David Young, Ohio University
Yunfeng Huang, Ohio University
Stephen Farrell, Trinity College Dublin
Joerg Ott, Helsinki University of Technology
Andrei Gurtov, University of Helsinki
Jeff Bush, MITRE
Prithwish Basu, BBN Technologies
Jim Snyder, DARPA/STA
Ryan Metzger, Lehigh University
Mike Demmer, UC Berkeley
Kevin Fall, Intel Berkeley Research Lab

I framed the question as follows: there are a number of different 
reasons for having a reference implementation. It might be (a) as a 
benchmark for interoperability with alternate implementations; (b) a 
platform for experimenting with the technology, e.g. trying out 
different routers and convergence layers, or variants of the protocol; 
and (c) a platform for people who want to build DTNs, a black box they 
can download and deploy in the field. What do we want the DTN RI to be?

I pointed out that the relative sizes of the groups were | (a) | << | 
(b) | << | (c) |, i.e. there would be a handful of people building 
alternate implementations; there might be tens or hundreds interested in 
experimenting with routing or the protocol; there would be (eventually, 
we hope) thousands or hundreds of thousands of DTNs deployed.

As a point of comparison, there are a (relatively) small number of HTTP 
daemons; a larger number of people who implement Apache plugins; a much, 
much larger number of people who deploy Apache out-of-the-box, only 
modifying config scripts.

I took away that people want (b), a platform for experimenting with DTN 
itself. It was pointed out that if (when) the need for easily-deployed, 
high-performance and/or high-reliability DTN implementations arose, it 
would be met by the commercial world. (The need might be met by taking 
the reference implementation and modifying it, or it might be met by 
clean-slate implementations -- but that wasn't the goal for the DTNRG RI.)

Mike Demmer told me that he plans to finish his Ph.D. this summer, and 
is thinking that he's interested in finding a start-up, which will leave 
him with little time to maintain DTN2. He hopes to be able to spend some 
time on it, given his investment in the code, but the time is likely to 
be very limited.

Someone suggested that Ohio University might be a good place for the 
code and DTNRG server to call home, long-term. Other suggestions that 
came up were Boston University and Google (Vint Cerf owns the domain 
name dtnrg.org; he might be able to get his employer to maintain the 
DTNRG server).

Ohio University might also be a good place for an interoperability lab, 
assuming funding could be secured.

Someone pointed out that the RI had done quite well being maintained by 
a graduate student, and perhaps what we should do is find another grad 
student to take it on as a project.

Stephen Farrell espoused the position that there was no need for a 
single distinguished reference implementation. He pointed to the Jabber 
community, where there are dozens of servers and clients available, and 
a well-specified protocol. He suggested that instead of having one 
implementation listed as the RI on the web page, list any 
implementations available, with comments describing their attributes -- 
this one is written in Java, this one is small and lightweight, this one 
scales, this one is good for hacking on, this one is easy to deploy.

I apologize for my notes being incomplete; we met at a pub, and it was 
difficult to hear the person next to you, much less the people at the 
other end of the table. If anyone has additional notes they are willing 
to share, I would appreciate them posting them here.

My thanks to Jim Snyder for co-sponsoring the beer.

- Chris

-- 
Dr. Christopher Small                         617.873.6261 (vox)
Networking Research                           617.873.6091 (fax)
BBN Technologies | MS 6/5C | 10 Moulton St | Cambridge MA, 02138