Accelerating Sketch-based Network Flow Processing Using Graphics Processing Unit

摘要

Sketch-based methods are widely used in highspeed network monitoring applications. In this paper, we present a parallel implementation of sketch computations using Open Computing Language (OpenCL) on Graphics Processing Unit (GPU) for network traffic change detection. The parallel nature of the computation on sketch data structure makes it suitable for implementation on GPU. Comparing to the implementation on CPU, the experiment results show the speed up of 2.8 times and 4 times in average for sketch table size of 1024 up to 32768 using two different forecasting methods.