Abstract
In this thesis, we describe a two-phase approach for cost reduction on Bluetooth inter-piconet communication. A cost model is established on the access pattern of bluetooth devices during a communication session, which performs a profiling mechanism. Due to the physical limits on Bluetooth specifications, inter-piconet communications are cost expensive, and we can not put lots of nodes in the same piconet. The end-to-end message cost between each node pair is determined by the number of hops on nodes. We apply a bipartite topology manner for constructing a initial scatternet and reconstructing the initial one. Traditional graph partitioning scheme is used to help us to distinguish what nodes should be grouped to a cluster. We adopt a recursive graph bisection method and KL refinement procedure to achieve nodes grouping using access frequency. These balanced clusters have to connect by the role of nodes for forming the last scatternet topology. Applying weighted bipartite matching method to assign the appropriate node attributes with the result that we configure a low message cost scatternet. Our experiments are performed on simulating the variation of total message cost in adjusting diverse parameters. In variables with uniform distribution, we improve the original cost about 13%. If choosing a test set which owns natural group property, we can obtain the better improvements. In a distributed computing model such as Java RMI, our works can greatly enhance the transmission cost between method invocations. This facilitates the distributed computation applications to implement on Bluetooth environment.