[e2e] TCP smarm

[Jon Crowcroft]

In missive <200404261635.JAA07867 at gra.isi.edu>, Bob Braden typed:

 >>I would like to ask each of you to limit your contributions to the
 >>current threads to a maximum of one message to the list per 24 hour
 >>period, per thread.
 
 Bob,

ok - here's a new thread.

TCP encouters a lot of problems when confronted with random loss -
packets in flight on wireless links have a higher chance of loss due
to turbulance - little vortices in the ether are setup as the edges of
the bits in the packet interfere. Its clear that there are several
solutions that involve changing TCP, but I intend to experiment with a
sub-IP solution, but still an end-to-end (or at least, wall-to-wall)
solution.

What we need to do is to reduce the fuzziness aroudn the edge of
wireless packets, so that they travel more smoothly through the ether
- this calls for a balanced asynchronous encoding of packets,
something we call Forward Edge Completion - it can be implemented in
the link layer, or in IP, but it can also easily be done by a new TCP
option, which encodes the payload differently - essentially we borrow
the idea from NRZ(I) type frame encoding techniques -if you look
closely at a typical TCP data packet, there are lots of 1s and 0s -
the 1s interfere much more than 0s, so what we need to do is to
runlength encode the values as 
n 0s
is the value n

clearly there's a tradeoff, as the packets get longer, and even 0s can
interfere. so we need to workout what the level or turbulance in the
ether is - this is hard - some call this an oracle - I think of it as
an adversarial priest (a bit like maxwell's demon) who is trying to
mess with transmission and also trying to hide their existence - but
we can use information theory to figure out how adversarial the
turbulance priest is, and then rid ourselves of them by the use of the
FEC with an appropriate redundancy levels - lets refer to these as:
Non Interfereing Transmission Entities, NITEs - I reckon, looking at
the weather out there, that about 4 NITES should be sufficient to rid
us of 1 turbulent priest.


 cheers

   jon