ANLS: Adaptive Non-Linear Sampling Method for Accurate Flow Size Measurement

Sampling technology has been widely deployed in network measurement systems to control memory consumption and processing overhead. However, most of the existing methods suffer from large errors for the estimation of small-size flows. To address this problem, we propose an adaptive non-linear sampling (ANLS) method for flow size estimation. Instead of statically pre-configuring the sampling rate, ANLS dynamically adjusts the sampling rate for each flow according to the value of a corresponding counter. A smaller sampling rate is utilized when the counter value is large, while a larger sampling rate is employed for a smaller counter. In this paper, the unbiased flow size estimation, the relative error, and the required counter size are studied through theoretical analysis and experimental evaluations. The analysis and experiments demonstrate that ANLS can significantly improve the estimation accuracy (particularly for small-size flows), and save memory consumption, while maintaining processing overhead comparable to existing methods. Moreover, we validate the design of ANLS by implementing an FPGA-based prototype, which is capable of measuring traffic throughput up to 26.5 Gbps.