[e2e] What should e2e protocols know about lower layers?
Panos GEVROS
P.Gevros at cs.ucl.ac.uk
Thu Oct 11 03:53:43 PDT 2001
RJ Atkinson typed :
|As I also read the TSVWG list, I'll note that I think that congestion
|avoidance and control mechanisms should always be turned on.
|I disagree with folk disabling congestion avoidance and control
just because a given link might be perceived to be "local"
|(particularly where "local" seems to have the de facto meaning
|of "on-subnet").
although disabling congestion avoidance and control is hard to justify and
probably not a wise thing to do in any case, the ability of the transport to
*relax* (make more aggressive) the congestion control rules ( or negotiate
what is the appropriate congestion control behaviour based on information the
endpoints have e.g. policy or some defintion of "locality") may enhance
performance, increase utilisation, benefit the application and the network
provider in general
so overriding congestion control was probably too radical a measure to deal
with locality , imho adopting a more aggressive behaviour (but still
controlled) may be justified
the counter-argument is that diversity in the congestion control behaviours
may well create a mess which will shadow the performance/other benefits . but
this remains to be seen.
cheers
Panos
More information about the end2end-interest
mailing list