|
Improved Event Processing Performance through Parallel Event Transformation Appeared in: The Second International Workshop on Event-driven Architecture, Processing and Systems (EDA-PS'07) Date: 2007 Download: BIBTEX (brito2007edaps) Applications like data processing from RFID, sensor networks, algorithmic trading and network management are intuitively mapped in a Event-Driven Architectures (EDA), but have requirements (i.e., throughput, latency, search for complex patterns) that were not available in traditional event-driven systems. Recently, these requirements have been satisfied by technologies like Complex Event Processing (CEP) and Event Stream Processing (ESP). In this work, we investigate a general-purpose CEP/ESP tool named Firefly and propose solutions for increasing parallelism in event processing. In particular, we focus at the transformation phase of events, which is responsible for executing computation over the events before they can be processed by a correlation engine. We show an approach that statically analyzes rules and that groups them according to their event dependencies. Through this approach we extract available parallelism while still respecting the two correctness requirements: (1) the order of the events in the stream is preserved; and, (2) if there are dependencies between these events, they are processed sequentially. |