From: Kate Everitt (kteveritt@yahoo.com)
Date: Sun Oct 17 2004 - 22:38:01 PDT
Paper: Random Early Detection Gateways for Congestion
Avoidance
Reviewer: Katherine Everitt
This paper discussed using maintaining average queue
size as a technique for congestion avoidance. The
system it uses to mark or drop packets in order to
notify a connection of congestion is designed to be
more fair than previous schemes. It drops packets
relative to the related connection’s bandwidth use.
This has the advantage that connections with bursty
traffic are not penalized by being more likely to get
dropped. The other main advantage of this scheme is it
is less likely to get synchronization effects which
cause everyone to back off at the same time because of
congestion.
The most interesting idea I found in this paper was
the difficulty in implementing a congestion scheme
which is appropriate for all sorts of traffic. If the
domain is known, it is a lot easier to come up with an
optimal algorithm. This difficulty is compounded by
the changing capabilities of the hardware. For
example, one previous scheme DECbit was seemed no
longer appropriate because of changing network speed
and buffer size.
Another concern I had was that the implementers did
not want to impose a penalty on bursty traffic.
However, this kind of traffic can be disruptive to the
equilibrium of the network and akin to misbehavior,
and the canned simulations in the paper did not
convince me that always supporting it is the right
approach.
This paper was a very thorough exploration of the
random early detection gateways, and they appear to be
an effective tool in protecting queue sizes and
network equilibrium, to avoid congestion. I would have
liked to see more realistic simulations, and am
concerned about how this system would scale given the
importance of selecting appropriate parameters.
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
This archive was generated by hypermail 2.1.6 : Sun Oct 17 2004 - 22:38:07 PDT