From: Seth Cooper (scooper@cs.washington.edu)
Date: Wed Dec 01 2004 - 02:24:12 PST
This paper presents a method for controlling ill-behaved aggregates in
the Internet. An aggregate is considered to be a group of flows that
share some common characteristic - most commonly, the destination
address. An aggregate that is consuming more that its fair share of
bandwidth is possibly a denial of service attack or a flash crowd that
should be controlled. The paper discusses how to perform
aggregate-based congestion control (ACC). This involves a router
determining if there is an aggregate that is congesting it and deciding
how to limit it. The router can perform merely local ACC itself, or may
perform "pushback", there it enlists the help of upstream routers to
drop packets that it would have dropped anyway.
One strength of the algorithm presented is the ability for the ISP
controlling particular routers to apply its own policy. For example, an
ISP could determine how to decide when a router should decide that it is
congested, how much it should limit the rate of the congesting
aggregate, and whether it should use pushback. This gives the people
running the ISP more control over how aggregates will be controlled by
ACC. This in turn makes it more likely for ISPs to adopt ACC in their
networks and gives a way to slowly deploy it throughout the Internet.
A problem with the algorithm is what the paper terms "collateral
damage". If traffic from a well behaved flow too closely matches an
aggregate that is causing congestion, it may have its packets dropped
just as if it were actually part of the aggregate. The process for
finding addresses to identify aggregates may be too general; it might be
better to err on the side of caution so that good flows are not
mistakenly punished. Also, it is not entirely clear how well pushback
would work across interdomain links. If a router is being congested by
an aggregate from another ISP, it is not clear that the other ISP has
anything to gain from responding to the pushback messages and limits the
aggregates rate on its end.
This paper is important because it addresses the issue of protecting
the Internet from attacks. The Internet is a very open network that
currently has little protection against attacks from malicious
aggregates that cause congestion. Denial of service attacks are nothing
new; yet they are still a problem for which there is no sure solution
deployed. This paper presents the beginnings of what could be a good
solution to denial of service attacks and other aggregate-related
problems in the Internet.
This archive was generated by hypermail 2.1.6 : Wed Dec 01 2004 - 02:24:18 PST