Make filters smart in multimedia streams environments

We consider the problem of optimizing and executing multiple continuous filtering rules, where each filtering rule is a conjunction of filtering units and each filtering unit may occur in multiple filtering rules. We focus on the problem of ordering the filtering units adaptively to minimize the cost in a multimedia stream environment where filtering units are very expensive. Significant performance gains are achieved by sharing filtering units across filtering rules. Existing methods are based on a greedy method which orders the filtering units according to three factors of the filtering units, i,e., the selectivity, connectivity, and cost. Although all these methods reported good results, there is still one important problem that hasn't been addressed yet. The selectivity factor is set by prior-knowledge which cannot adjust to the multimedia stream where characteristics of stream and filtering units vary unpredictably over time. In this paper, we propose a hierarchal clustering ordering framework (HCOF), which executes a two-stage ordering strategy. In the first ordering stage, HCOF clusters all the filtering units according to their popularities and costs factors. In the second stage, for each cluster, all the filtering units are then ordered according to their selectivity. Experiments on both synthetic and real multimedia streams demonstrate that our HCOF method outperforms other existing filtering methods.