Cell Relay Archive

["Jim Sackman" <jsackman@pacbell.net>]

<snip>

    Yes, I am missing the point.  The problem is it either works enough for
people to pay money for or not.  Lets use VoIP as an example.  Net2Phone
works well enough because its free.  Suppose you were paying standard long
distance rates?

    Actually, end stations (read Microsoft NetMeeting) do support RSVP.
Routers do not, because it requires them to support differentiated IP
services (gee, like ATM) which are complicated and expensive.  So, backing
down in the implementation chain would be DiffServ.  Implementing that
protocol would be much more like what you suggest.  It does not create real
QOS, but makes higher priority traffic work better (better yet to be
defined).  In a similar vein, MPLS will help.

    All these IP QOS mechanisms, pale in front of ATMs capability.  ATM does
not provide the end to end services of IP, but that is why IP over ATM is
one of the standards in the Internet backbone.  The problem is that we have
no effective way of signalling differentiated services.  As the bulk of the
traffic on the Internet is the equivalent of UBR, we could use UBR PVCs as
trunking between routers.  My suggestion was to have the end stations signal
for "special" treatment of bandwidth using the only protocol available
(albeit not implemented in the routers).  This would allow the special
traffic to be segmented at the edge and treated specifically (and billed
separately from the carrier's perspective).  Alternately, this could be done
by have a TCP server proxy services at the exit of a premise (I am assuming
that the premise has, in practice, unlimited bandwidth).  This TCP service
broker could then provide a separate connection (ATM SVC) for the
specialized service.  It seems to me that using the QOS protocol we have and
making it effective with the networking protocol we have is the shortest
approach to the goal line.


--
Jim Sackman
Chief Technologist
Advanced Fibre Communications