Relatively effective and practical load shedding strategy for sliding-window join queries over data streams

We now present a new strategy quite effective and practical for shedding load from sliding-window join queries over data streams. We propose the problem description, sliding-window join queries, a load shedding strategy based on the partition of the domain of join attributes. The schematic shows how to execute the strategy with two operator modules X1 and X2. The strategy is essentially that the domain of the join attributes is partitioned into certain sub-domains, and tuples are dropped according to their join values by maintaining simple data stream statistics. We performed two experiments. One is concerned with the effect of different skew parameters of zipf distribution, the other is concerned with the effect of different overloadings. Results of experiments are shown in the paper. Our new strategy needs fewer statistics of input data streams and it makes it convenient to further process the outputs of join operation. It also has good adaptability for different skew parameters of zipf distribution and different peak loads. The theoretical analysis and experiments show preliminarily that the new load shedding strategy is effective and efficient for window join queries.