Space-efficient tracking of network-wide flow correlations

Abstract

The information of temporal correlations among network-wide data flows is crucial to a wide range of network management applications, such as root-cause analysis, threat monitoring, and traffic profiling. While several prior work had only studied the centralized and offline computation of flow correlations, we present DisTrack, a space-efficient network management mechanism for online tracking of network-wide temporal flow correlations. The major benefits of DisTrack include low space complexity, high processing speed, and ease of distributed deployment. This paper presents its randomized data structures, with theoretical analysis on the trade-off between space complexity and accuracy. We further provide extensive empirical evaluations on real network traces.