Giving control to end-systems
Proposals for an Internet with differentiated services are usually
based on a small number of service classes with well defined quality
and prices. However the approach has drawbacks
In particular, the development of ATM
traffic classes has illustrated some of the difficulties of
defining service categories before the applications that might use
these categories have been invented or have become widespread,
difficulties compounded by the requirement that associated pricing
schemes be incentive-compatible for users.
In this talk we describe a radically different
approach to differentiated services. Its premise
is that a simple packet network may be able to support an arbitrarily
differentiated set of services by conveying information on congestion
from the network to intelligent end-nodes, which themselves determine
what should be their demands on the packet network.
We argue that there would then
be no need for large buffers (and their associated latency)
or priority queues within the network, or
for connection acceptance control at the border of the network.