With so many badly written papers being published each month, it is difficult to be critical of one that is written so well, and so clearly. The only aspect of the paper available to criticize is its academic content, and it is a challenge to do so without inadvertently revealing one’s own ignorance.
This paper is like others in contemporary journals, discussing network queueing theories and rate 1 adversaries. The authors start by showing that there are packet sequences for which no scheduling rule can guarantee delivery of all packets within a finite time, with respect to rate 1 adversaries. The authors demonstrate this using a network containing cycles of length greater than two.
The discussion then moves on to sequences that allow bounded delivery times, where the authors propose a protocol called estimated-rare first. This protocol determines which links are rarely used, and prioritizes all packets that traverse these links. In cases where there are no prioritized packets queued at a node, the remaining packets are sent using the furthest-to-go rule. The protocol also uses control packets to propagate information regarding routes that have been estimated and subsequently marked as rare. The protocol, however, hinges on the adversary being periodic, rather than dynamic, as is more likely in practice.
With a background as a protocol practitioner, I find that current academic research has digressed so far from practice that it has become difficult to drive knowledge from the former to the latter. In many respects, the dominance of best effort protocols is a testament to this. Sadly, real networks have more to deal with than controlled, rate 1 adversaries, and instantaneous transmission links.