This paper presents a new packet scheduling algorithm called Stratified Round Robin. It combined timestamp scheme and the Round Robin scheme to achieve both good performance and low complexity. The main contribution of this paper is its easy and straightforward hardware implementation and convencing O(1) complexity. The authors also differentiated their algorithm from other algorithms with similar complexityby emphasizing that only they achieved a signle packet delay bound that is independent of the number of flows.

Most of us agreed that the paper did a good job in explaining the algorithm and providing detailed and thorough analysis. However, the simulation part of this paper is weak. Firstly, it only compared Stratified RR with DRR and WFQ, two very old algorithms published more than a decade ago. Secondly, since Stratified RR algorithm follows the same approach as the other algorithms, such as Smoothed RR, Aliquem and BSFQ, it will be very interesting to see a fair comparison among them. Finally, the simulation setup is very simple, so we really don't get much from the simulation results. A more comprehensive simulation and insightful discussion are what is missing in this paper. For a well-studied area like fair queueing, it is surprising to see that no standard testbench is available for a fair comparison among different algorithms.

The paper analyzed the worst case delay bound and the single packet delay bound. Since the Stratified RR uses the WDRR as its intra-class scheduler, it unavoidablly inherits the fact that the worst case delay bound is proportional to the number of flows, which all Round Robin based algorithms suffer. But the paper argued that their single packet delay bound is only a function of the reserved bandwidth. But note that the reserved bandwidth is still somewhat related to the number of flows. In this sense, there is not much to cheer about. Also Dr. Turner pointed out if the difference between any of the delay bounds they have achieved and the ideal GPS delay bounds is irrelevant with the number of flows and the reserved bandwidth, that will be a very strong argument. But seemingly they didn't succeed. Also, when we consider QoS guarantee which one of the two delay bounds is more important is still an open question. In congested network senario, a single packet delay bound seems not a useful property.

The Stratified Round Robin algorithm has a lot of similarities with the Smoothed Round Robin and Group Rate Round Robin algorithms. The paper cited them, but didn't give the merit they deserved. Actually none of the ideas from this paper is absolutely original. Stratified Round Robin just combined these ideas and made a practical algorithm that is easy to implement in hardware. But we also should admit in a well-studied area like fair queueing, it is very hard to come out a brand new algorithm. So we shall appreciate any improvement and contribution. At the end of the discussion, the audiences of this class voted and ranked this paper somewhere around 3 to 5.

Jing Lu