Long-term Smooth Per-Flow Size Measurement Mechanism for High-Speed Networks

Efficient measurement of the size of each flow in high-speed networks can provide basic data support for practical applications such as network resource optimization and anomaly detection. However, most of the existing work on per-flow size measurement only focuses on a single measurement period, or multi-period cumulative size information measurement, while lacking the tracking of the long-term dynamic change trend of the per-flow size. To address this problem, the concept of smooth per-flow measurement is introduced, and the SmoothSketch mechanism is proposed to achieve long-term and controllable tracking and measurement of per-flow size. The designed mechanism first separates large and small flows based on the Count-Min structure, and further dynamically stores and maintains the smooth per-flow size information of large flows based on the bucket array and cuckoo hash strategy, thereby achieving low-overhead and high-precision smooth per-flow size monitoring. Experimental results based on real network datasets show that, compared to baseline algorithms, SmoothSketch’s Average Absolute Error and Average Relative Error are reduced by 26.83 times and 38.68 times, respectively, while its query throughput is increased by 43.54 times.