Summary: Analysis and Simulation of a Fair Queuing Algorithm
This paper presents a fair queuing algorithm which is based on a previous work done by Nagle. The
aim here is to provide some fairness in bandwidth and delays observed by flows and to provide
protection from ill-behaved sources.
The basic idea behind the algorithm proposed in this paper is to closely simulate a bit-by-bit round
robin scheduling algorithm for packet switching. The algorithm differs from Nagle's proposal in that
they take into account the packet lengths as well. The authors also take into account the allocation of
bandwidth, buffer space and delay and also separate delay allocation from the other two.
For analyzing the algorithm, the authors assume two arrival streams- FTP like sources and a telnet-like
interactive source. With this interaction, they show that (1) FQ provides lower delay to lower
throughput sources independent of flow control at end points and (2) No conversation gets more than
its fair share; queuing delay diverges with throughput relative to fair share approaching 1.
The authors also provide a variety of simulation results in which they measure throughput, RTT,
number of retransmissions and dropped packets with different simplistic topologies and with different
combinations of flow control algorithms at end hosts. They simulate an underloaded gateway,
overloaded gateway, ill-behaved source, mixed protocols, multihop path and complicated network.
Overall the results show that FQ delivers satisfactory congestion control in all scenarios.
In my opinion, while the proposal is very significant, its effectiveness hasn't been fully explored or
proven in the paper. The simulations are done on simplified topologies, typically to illuminate the
features of the algorithm and a detailed measurement study on the Internet would be more satisfying.