From: Daniel Lowd (lowd@cs.washington.edu)
Date: Mon Oct 25 2004 - 01:57:42 PDT
This paper introduced and analyzed a queueing algorithm that met certain
fairness criteria, specifically that users who used less than their
allotted bandwith would receive faster service, and that no user could use
more than its portion. The main advantages of the algorithm over previous
work seem to be that it takes into account packet size and that it can
offer lower latency to intermittent conversations. The analysis seemed
fairly involved; it's difficult to test such a change in every possible
scenario. They seem to have chosen a number that illustrate the potential
advantages of their method.
The biggest problem with the algorithm not that it seems wrong, or flawed,
or limited... but that it seems obvious. Once you decide that round-robin
rewards large packets, bit-by-bit round-robin is the logical remedy... and
to create something that's equivalent ot bit-by-bit round-robin but
feasible... well, that's just bookkeeping and algebra, right? The \delta
parameter is a cute hack that gives an advantage to intermittent
conversations, but the rest was fairly straightforward. And then, to fill
up the space, they put in lots of formulas and tables and numbers. Do I
really care if "A(P) = N(P - P_F/2) for P \geq P_F"?
In conclusion, I think that this paper presents a valuable algorithm along
with a fair bit of useful analysis. The idea presented is very simple --
it just makes sense. If only the analyses and experiments were the
same...
This archive was generated by hypermail 2.1.6 : Mon Oct 25 2004 - 01:57:43 PDT