A cost model for complex event operators
Umesh Bellur · 2014
Complex event detection is both CPU and memory intensive and is difficult to scale to process the high event-rates. One way of dealing with high resource costs is to distribute the work of a Complex event detector. Distribution must happen along the axes of different operators making up a complex query. This in turn needs a good understanding of the "cost" of a operator. In this paper we present an analytical cost model of complex event operators, that can serve as building block for distribution and placement of queries in distributed complex event processing. The cost model estimates CPU and memory requirements using input event-rates and query parameters like predicate selectivity and window-size. We have verified the accuracy of this cost-model using a Java based implementation of a CEP detector and with real world event streams (but scrubbed for specific predicate selectivities) and show that the prediction is accurate to within 5% of measured values.