Re: [dtn-interest] dtn question

[Scott Burleigh <[e-mail]>]

At 08:54 AM 12/15/2004, magical monkey wrote:
>Hi,
>
>I'm a senior at Yorktown High School doing a science project related
>to the efficiency of a delay-tolerant-network.  I was wondering if I
>could get some feedback on whether the project seems reasonable or
>useful - I'm planning to study the delay in transfer by varying delay
>times from 1000-60000 seconds, in intervals of 1000 to study the
>transfer success ratio and overall transfer time as delay is increased
>(and do statistical analysis on the data).  Any comments or
>suggestions would be greatly appreciated!  Thanks a lot.
>
>Sincerely,
>Anna

Hi, Anna.

I believe this will be pretty challenging in the near term, as we don't yet 
have an implementation of a "convergence layer" adapter to a protocol that 
can efficiently accomplish reliable data transfer between two points 
separated by the kind of delay you're talking about.

If you're looking at a 1000-second delay end-to-end but brief delays 
point-to-point (for example: A sends to B via TCP convergence layer over an 
Ethernet at a time when B has no connectivity to C; you wait 1000 seconds, 
then break the A-B connectivity and establish B-C connectivity, and B then 
sends to C via the TCP convergence layer over an Ethernet; you've got a 
1000-second end-to-end delay between A and C), then it's doable with the 
code we've currently got.  But in that case I'd be concerned that you 
wouldn't get very interesting transfer success and overall transfer time 
statistics.

I suspect that the sort of study you want to perform will require 
implementation of a convergence layer adapter to something like the 
Licklider Transmission Protocol (LTP), which does reliable point-to-point 
transmission over very long delays.  Stephen Farrell at Trinity College 
Dublin has been working on an implementation of LTP itself, but for the CL 
adapter I'm afraid you'd be on your own.  Not a trivial project.

Scott