From: Rosalia Tungaraza (rltungar@u.washington.edu)
Date: Wed Oct 27 2004 - 00:12:01 PDT
This paper focuses on modifying the "best effort" packet delivery service
that the Internet offers to better assure users (who are willing to pay
the price) that their packets will be delivered. The method proposed by
the authors is based on marking packets as either in or out depending on
whether or not the user has paid for the guaranteed service. The former
user has his/her packets placed in a specific profile, from which every
packet has a very high chance of getting delivered to its destination in
time of congestion. A packet marked "out" on the other hand, has a high
probability of being dropped during a congestion. I use the words "high
chance" for packets marked "in" because in cases where the routers are
flooded with "in" packets and there is congestion, some of those packets
will have to be dropped despite the fact that they are "in" packets.
Profile meters placed at the edges of the networks monitor which packets
get dropped or propagated to the next network/host. The location of these
meters depends on whether the network has adapted a receiver-based,
sender-based, or a combination of those two mechanisms. Moreover, the
authors propose the use of the RIO algorithm to determine packets to be
dropped. The RIO algorithm is a modification of the RED algorithm in such
a way that the in and out packets are given different probabilities of
being dropped. Out packets have a much higher probability compared to in
packets.
One success of this paper and thus the algorithm is that the algorithm
allows users to choose what priority they can send their packets with a
relatively high guarantee that their high priority packets will reach
their destinations. This is not the case for the "best effort" scheme
where the Internet tries its best to transport a user's packet but can
never provide any assurance of whether the packet will actually be
delivered to its destination. Moreover, unlike the current Internet, this
algorithm provides a means of dealing with users that do not respond to
warnings about reducing the number of packets they are sending. For such
users, their packets are among the "out" packets and thus are continually
(preferentially) dropped during congestion.
One idea I would have liked them to expound is how cost-effective they
think it would be to change the current network to suit their proposed
algorithm. From my understanding of the paper, the major changes are
proposed to be only in major routers within ISPs. However, I wonder about
the various software/hardware changes that need to take place among
hosts/users. Would the present Internet be likely to adapt to this
algorithm?
As the authors suggest, one possible future work would be to implement and
test this algorithm with a real testbed.
This archive was generated by hypermail 2.1.6 : Wed Oct 27 2004 - 00:12:03 PDT