The IP over ATM Mailing List Archive by date

[Jon Crowcroft <J.Crowcroft@cs.ucl.ac.uk>]

 >>this is what i have been thinking too as a possible solution.  we             
 >>already have switches that keep state when they do early packet discard.      
 >>with similar logic it would be possible to schedule cells out from those      
 >>incoming vcis that already have full frames in the (shared memory)            
 >>buffer.  that would allow implementing mp-p without an application (=ip)      
 >>layer multicast server.juha                                                   
 >   As have been discussed a number of times before, this is the                
 >   most trivial way of immitating IP framed based features over                
 >   ATM and simplest to implement. Solves a lot of problems and                 
 >   has minor disadvantages only. Can also be done at the multicast             
 >   servers very easily solving more problems than it introduces.               

of course, the difference between hunting for the bit and discarding,
and hunting for the bit before forwarding a "frame's worth of cells"
is that the former fixes congestion, whilst the latter removes the cut
through store and forward  advantage of forwardign a cell at a
time....but i guess since it only happens at a merge point for mp-p
conenctions, it only ever introduces 1 of these frame
store-and-forward serialisation times, rather than n for n hops...


 jon