Two-Stage Adaptive Bloom Filters For Per-Flow Monitoring In Software Defined Networks

Per-flow monitoring attracts great attention for its importance in networks. Thus a two-stage Bloom Filter is proposed for it. However, without considering the scalability of the two-stage Bloom Filter to different probability distributions of target flows' monitoring action types with high probability deviations, it leads to some serious problems, such as high false positive rate or low resource utilization. Therefore we propose the two-stage adaptive Bloom Filters that are operated collaboratively in network-wide range in software defined networks. The Bloom Filter supports dynamic mappings from the pre-undetermined action types of its constituent Bloom Filters to specific ones and adjustable number of the constituent Bloom Filters. This proposal has two major advantages: 1) false positive rate can be kept under a threshold in the different probability distributions; 2) resource utilization and rejection probability can be significantly improved at low cost of increasing the threshold. We analyze and discuss the two-stage adaptive Bloom Filters from false positive rate, resource utilization and rejection probability. The results from our simulation based on real-life network topology agree with our analysis and discussion.