BatchSketch
Wendi Feng, Chuanchang Liu, Junliang Chen · Proceedings of the 28th Annual International Conference on Mobile Computing And Networking · 2022
Heavy flow identification is essential for discovering potential adversarial activities in mobile edge networks. However, state-of-the-art falls short in accuracy with approximation algorithms and unbounded memory usages with precise measurement. To this end, we introduce BatchSketch, a "network - server" aligned solution to achieve both high accuracy and low memory usage. The intelligence behind is that BatchSketch first conducts coarse-grained filtering from the switch with bounded memory and computation resources, and it then sends the filtered flows to the RDMA-link attached server with plenty of memory for accurate measurement. Our primary experimental results indicate that the filter on the switch can filter 99% of non-heavy flows, remarkably reducing the memory usage for the measurement.