Paper Review: End-to-End Packet Delay and Loss Behavior in the Internet
Reviewer: Mark Meras (mm446)
Main Contribution
This paper by J. Bolot presents a new analysis of end-to-end packet delay
and loss behavior by using a "ping"-like mechanism. Bolot confirms
previous work on mixing high-bandwidth, large packets with small,
interactivity-intended packets. He also claims that the loss of these
"ping" packets (called probes) is random in most situations.
Key Ideas
- We need to analyze packet delay and loss behavior (i.e. how and when
packets are dropped). Packet delay includes both fixed delay (e.g.
propagation delay) and variable delay due to routing and queuing delay.
Loss behavior needs to be measured as to its frequency and clustering
- Analysis was performed using small "probe" UDP packets. The two
streams of probe packets and regular Internet packets were mixed together.
Analysis revealed that probe packet loss tended to be in bursts.
- Other, not so expected, results also occured. Probe packet loss was
random unless traffic intensity was very high. Rapid fluctuations of
queuing delays over small periods were also common.
Critique of Contribution
Most of the paper does not seem suprising. Clustering of packets during
loss has been studied previously. However, the randomness of packet loss
in low-intensity situation is an interesting update to the solution. More
importantly, it was shown that probe tools can be used to study end-to-end
network characteristics.
Open Question
The question remaining is how will multimedia transmissions (such as
real-time video) affect the nature of packet loss and delay on the
Internet? This is yet to be studied at the time of publication.