From: Kate Everitt (everitt@cs.washington.edu)
Date: Tue Oct 26 2004 - 18:37:03 PDT
D. Clark and W. Fang, "Explicit Allocation of Best-Effort Packet
Delivery Service," IEEE/ACM Transactions on Networking 6(4), August 1998.
Review: Katherine Everitt
This paper presents an “allocated capacity” framework which provides
differential services with a much simpler scheme than reservations,
namely marking packets “in” or “out” depending on the service allocation
profile of the endpoint. If a sender sends at a greater rate than it is
entitled to, the extra packets will be marked as “out” and will be more
likely to be dropped in times of congestion. This scheme is known as
RIO, which is random early drop with the in/out bit. One big advantage
of this scheme is that it doesn’t require a lot of complicated router
control of various queues (eg like round robin). All the packets can go
in one queue and still provide varying QoS. The probability of dropping
an in packet will depend on the average queue length of the in packets,
while the probability of dropping an out packet depends on the total
queue length.
One of the important issues not addressed by this paper is how the
endpoints will specify what their rate will be. This system is difficult
to implement in the Internet at large, because it will require a lot of
changes to routers. Also, the QoS guarantees seem iffy, because if
everyone wants to send data, how fair is it to say that later traffic
flows can have nothing (or all “out” packets), because you have already
committed your full capacity. This seems like a more realstic scheme
than some of the others we have looked at, but still will require a
concerted effort to intergrate into the current architecture.
This archive was generated by hypermail 2.1.6 : Tue Oct 26 2004 - 18:37:08 PDT