Re: TCP and ICMP soft-errors in IPv4/6 environments

[Pekka Savola <pekkas@netcore.fi>]

On Thu, 30 Oct 2003, Bill Sommerfeld wrote:
> > Has this been discussed in the transport area?  Is there something to be 
> > done about this, e.g. a BCP document discussing this practice and 
> > suggesting ICMP DU's are OK when establishing connections -- or something 
> > else?
> 
> Maybe the right answer here is a "slewed parallel" connection
> establishment; start with one address, and if you get a soft error or
> a RTO-sized timeout, try a second, third, fourth, etc., connection,
> but leave the prior ones open in the event they complete successfully.
> 
> Soft errors wouldn't kill the first connection -- they'd just be a a
> hint to try another connection in parallel.
> 
> This should probably be buried in a library routine so that
> applications don't have to reinvent this wheel.

This is probably an obvious first thought, as what would be the desirable
behaviour.  But what would that imply in practice?  That at least some
parts of the basic API functions like connect() be augmented by a set of
new functions for managing connectivity?  That could be a forever-project,
while what we'd like is to get a fix that'd be usable in this decade :-).

I'm not saying that's not a probably relevant option to consider, but I 
think folks have been thinking about replacing APIs before and not much 
has come out of it.. so I'm a bit reluctant to start on that particular 
path...

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings