From: Kevin Wampler (wampler@cs.washington.edu)
Date: Wed Oct 27 2004 - 00:24:20 PDT
A commonly mentioned shortcoming of the current structure of the Internet
is the inability to provide different levels of service to different
users. One method for addressing this is proposed in "Explicit Allocation
of Best-Effort Packet Delivery Service".
Perhaps the most obvious way of providing quality of service is to mark
packets with a priority level (one level per degree of quality guaranteed)
and to manage the packets from each level differently (such as in a
priority queue). This method has the disadvantage that it is both
somewhat involved in its implementation and because of the reliance to
some fixed number of service qualities may not be easily extended to meet
future network demands.
The scheme proposed in the paper addresses these problems by using a two
part scheme. First, at entry to or exit from a network packets are tagged
as either being withing a user's quality plan of outside of it. Routers
along the path then merely give preference to the delivery of the `in'
packets over the `out' packets. This allows different schemes for quality
assurance to be implemented by changing the policy for determining if a
packet is `in' or `out', but allows the routers to be unaltered. The
paper also gives more in depth details on the implementation and
simulations indicating that it provides improvements over current methods.
The advantages of simple implementation and flexibility and notable with
the method presented in this paper. My intuition, however, it that it
occupies an uneasy middle ground where it is just far enough from the
current system. Even the minor change of having routers respond to `in'
or `out' packets would require changing current routers. I suspect that
in reality the additional cost to implementing support for a quality of
service method which is more involved and more independent from current
protocols is probably small compared to the cost of changing all the
routers in a network in the first place. Thus it seems as if some of the
schemes they propose, particularly those for congestion control are
constructed to be too close to something which could be implemented in
today's network than is really necessary.
This archive was generated by hypermail 2.1.6 : Wed Oct 27 2004 - 00:24:20 PDT