Towards Approximate Event Processing in a Large-Scale Content-Based Network

Abstract

Event matching is a critical component of large-scale content-based publish/subscribe systems. However, most existing methods suffer from a dramatic performance degradation when the system scales up. In this paper, we present TAMA (Table Match), a highly efficient content-based event matching and forwarding engine. We consider range-based attribute constraints that are widely used in real-world applications. TAMA employs approximate matching to provide fast event matching against an enormous amount of subscriptions. To this end, TAMA uses a hierarchical indexing table to store subscriptions. Event matching in TAMA becomes the query to this table, which is substantially faster than traditional methods. In addition, the false positive rate of matching events in TAMA can be adjusted by tuning the size of the matching table, which makes TAMA favorable in practice. We implement TAMA as a forwarding component in Siena and conduct extensive experiments with realistic settings. The results demonstrate that TAMA has a significantly faster event matching speed compared to existing methods, and only incurs a small fraction of false positives.