Dynamic Migration of Processing Elements for Optimized Query Execution in Event-Based Systems

This paper proposes a method for optimized placement of query processing elements in a distributed stream processing platform consisting of several computing nodes. We focus on the case that multiple users run different continuous Complex Event Processing (CEP) queries over various event streams. In times of increasing event frequency it may be required to migrate parts of the query processing elements to a new node. Our approach achieves a tradeoff between three dimensions: balancing the load among nodes, avoiding duplicate buffering of events, and minimizing the data transfer between nodes. Thereby, we also take one-time costs for migration of event buffers into account. We provide a detailed problem description, present a solution based on metaheuristic optimization, and evaluate different aspects of the problem in a Cloud Computing environment.