| Optimizing the aggregate throughput of 802.11 WLANs is an important challenge due to its increasing popularity and shared, finite spectrum. The common practice in optimization design is to address a problem domain, such as channel assignment or transmit power control, and test it across various environments, topologies and traffic rates. This approach, however, provides only a limited context for network administrators, who must decide not only the best combination, but also the best ordering of these independently derived and tested solutions.;This work presents a characterization of the combinatorial and ordering effects of algorithms from five optimization domains, including: channel assignment, association control, transmit power control, bit-rate adaptation and beam form selection. A centralized, measurement-driven system is proposed, implemented, and tested in a simulator and a field deployment to administer these optimization configurations in an outdoor 802.11b WLAN featuring multiple, beam forming access points. Aggregate throughput results are processed by a decision tree to classify optimization configurations into top and bottom tiers using pairwise ordering and algorithm selection attributes. The results demonstrate that: (1) the relative ranking of an algorithm from an optimization domain is dependent upon the combination and ordering in which it is applied and (2) the ordering of a set of algorithms is as significant to final performance as the combination selected. This not only provides a series of optimization design guidelines, but also implies that the design of optimization algorithms must be considered over a broad set of combinations and orderings to fully determine their real world efficacy. |
