Mr. Fusion: a programmable data fusion middleware subsystem with a tunable statistical profiling service

Abstract

Voting is the process of combining multiple replies from replicated servers into a single reply. Data fusion is similar to but more general than voting. In data fusion, the input sources are not necessarily replicated servers, hence the inputs exhibit greater variance. Data fusion is a fundamental building block in distributed systems. It occurs in diverse contexts such as consensus, sensor networks, intrusion detection, and hierarchical resource monitoring, among others. This paper describes Mr. Fusion, a framework that provides data fusion in middleware. The heart of Mr. Fusion is a Fusion Core module that provides mechanisms for programming a wide variety of data fusion algorithms. Another part is a Fusion Status Service that monitors low-level outputs from the Fusion Core and alerts subscribers to divergent values or timings. The implementation borrows techniques from data warehousing and data mining.