The publish/subscribe (pub/sub) interaction paradigm has
recently played a central role in the development of a large number of
information dissemination applications such as stock trading, traffic
information, news tickers, and electronic auctions. Its success is mainly attributed to its capability to decouple
interacting participants, which makes it a good candidate for the development
of such applications in mobile wireless environments that are characterized by
frequent and unpredictable disconnections of participants due to wireless
channel impairments or user mobility. Most current pub/sub middleware systems
are optimized for fixed networks (i.e., users do not roam and the
infrastructure itself is fixed). Therefore, add-on protocols are
needed to extend such systems to cope with the challenges imposed by
user mobility.
This thesis presents a novel and efficient mobile management scheme that is based on a pro-active caching approach (i.e., context transfer/caching occurs prior to the subscriber’s movement) to extend current pub/sub systems to support mobility. This approach is based on the use of a data structure called neighbor graph, which dynamically captures the set of next potential brokers to ensure that subscriber context remains always one hop (broker) ahead of its current broker. The proposed approach employs “dummy” subscribers that automate the task of context caching and removal at immediate neighboring brokers on behalf of the actual moving subscribers. We have extended a JMS-based pub/sub system with our pro-active caching approach and observed the incurred overhead of the approach. This is achieved by comparing the end-to-end latency of message delivery as well as the latency of message routing with and without enabling our pro-active caching approach.
We have comprehensively evaluated the effectiveness of our proposed approach through testbed experiments, comparing it to the state-of-the-art solutions, durable subscription-based and reactive, proposed in the literature. The experimental results show that our pro-active approach reduces the message loss by more than 50% and message duplication to zero, compared to durable subscription-based approaches. The results also indicate that our approach experiences much lower handoff latency compared to reactive approaches. Overall, our proposed approach achieves superior performance across a range of scenarios. We conclude our work by discussing a modeling approach that can be used to extrapolate the performance of our approach in a near-size environment (in terms of broker and/or subscriber population) to our experimental environment.