Review: End-to-End Internet Packet Dynamics

Review: End-to-End Internet Packet Dynamics

This paper discuss results from a large-scale measurement of the
internet packet dynamics. The measurement was conducted by tracing
20,000 TCP bulk transfers between 35 Internet sites. New findings
regarding unusual network event s such as out-of-order delivery,
packet replication and packet corruption are presented. It then
proposes a more robust procedure called "packet bunch modes" (PBM) to
estimate bandwidth bottlenecks of end-to-end communications. Finally,
the paper provides measurement findings about end-to-end packet loss
and packet delay.

It was found that out-of-order delivery in the Internet is fairly
prevalent. Data packets are more often reordered than acks because
they are frequently sent closer together, so their reordering
requires less of a difference in transit times. Reordering is also
highly asymmetric. A general conclusion drawn from the measurements
is that Internet paths are sometimes subject to a high incidence of
reordering, but the effect is strongly site-dependent, and apparently
correlated with route fluttering. Besides, while out-of-order
delivery can violate one's assumptions about the network that it is
well-modeled as a series of FIFO queueing servers, it often has
little impact on TCP performance. For packet replication, the paper
concludes that this is a rare phenomenon in the Internet. And for
packet corruption, the paper can not make make any useful conclusions
from the contradicting results obtained.

The paper then proposes a new bottleneck bandwidth estimation
algorithm called "packet bunch modes" (PBM) to compensate the
inefficiency of the original method based packet pair. Packet pair
works well on addressing out-of-order delivery, limitations due to
clock resolution, changes in bottleneck bandwidth, but fails to solve
the problem of multi-channel bottleneck links in a satisfactory way.

Finally, the paper presents the measurement results about packet loss
and delay in the Internet. As far as the packet loss is concerned,
the paper discusses how frequently the loss occurs, the differences
between loss rates of data packets and acks, the degree to which loss
occurs in bursts, and how well TCP retransmission matches genuine
loss. For packet delay, issues such as timing compression, queueing
time scales and available bandwidth have been discussed.

In all, the paper represents a solid piece of research work, based on
a large-scale measurement study. However, the paper's originality is
still lacking, probably because the topics touched in this paper have
all been well studied before. It sounds to me the only originality of
the paper is that it proposes to measure the available bandwidth in
bottleneck estimation. It is based on this key method that the issues
of packet loss and packet delay are revisited.
Received on Wed Nov 15 2006 - 21:54:36 EST