Accelerating Sketch-Based Computations with GPU: A Case Study for Network Traffic Change Detection

Abstract

Sketch-based algorithms are widely used in networking applications due to its many good attributes. We propose to use Graphics Processing Unit (GPU) as an accelerating engine to offload heavy sketch computations for network traffic change detection. Our experiment results show that GPU can conduct fast change detection with query operation up to 9 million distinct keys per second. It is capable of processing sketch data structure for wide-range of applications in fine-grained time scale efficiently.