Fair Queueing and Stride Scheduling
Fairness related objectives appear to have gained their prominence through fair queueing algorithms and stride scheduling — both essential building blocks of today's information technology. This chapter focusses on the fundamental issue of defining and quantifying fairness for these two applications rather than on the technical details of their implementation and performance which can be readily found in the literature, see for instance Keshav , Bertsekas and Gallager , and Waldspurger and Weihl .
We first observe that though both basic fair queueing and stride scheduling use essentially the same algorithms based on the Jefferson's method of apportionment their approach to defining and quantifying fairness have been rather different. While the former is based on the max—min criterion, and the relative as well as the absolute fairness bounds, the latter is based on the bottleneck and variance minimization. Despite these differences we propose here a common ground for both approaches to fairness in fair queueing and stride scheduling. This approach is based on the apportionment theory and the just-in-time optimization. It opens possibilities for alternative fair queueing and stride scheduling algorithms. One such clear alternative is the Webster's method of apportionment not previously used in either fair queueing or stride scheduling contexts. This method is the only one that results in a peer-to-peer fairness consistent with a standard two-client solution — a new and promising in the context of technical systems concept of fairness.
KeywordsBandwidth Allocation Outgoing Link Generalize Processor Share Fair Queueing Response Time Variability
Unable to display preview. Download preview PDF.