Accurate Per-Flow Measurement with Bloom Sketch

Sketch is a probabilistic data structure, and is widely used for per-flow measurement in network. The most common sketches are the CM sketch and its several variants. However, given a limited memory size, these sketches always significantly overestimate some flows, exhibiting poor accuracy. To address this issue, we proposed a novel sketch named the Bloom sketch, combining the sketch with the Bloom filter, another well-known probabilistic data structure widely used for membership queries. Extensive experiments based on real IP traces show that our Bloom sketch achieves up to 14.47× higher accuracy compared with the CM sketch, while exhibiting comparable insertion and query speed. Our source code is available at Github [1].