Economical and efficient network super points detection based on GPU

TL;DR

This paper presents a GPU-based algorithm for real-time detection of network super points that is cost-effective, requiring less memory and hardware resources, and capable of handling high-speed network traffic with high accuracy.

Contribution

The paper introduces a novel double direction hash function group that reduces memory usage and enables high-speed super point detection on low-cost GPUs.

Findings

Reduces memory usage to less than one-fourth of existing algorithms.

Capable of processing 750 Gb/s network traffic on a $200 GPU.

Achieves high accuracy in super point detection at high network speeds.

Abstract

Network super point is a kind of special host which plays an important role in network management and security. For a core network, detecting super points in real time is a burden task because it requires plenty computing resources to keep up with the high speed of packets. Previous works try to solve this problem by using expensive memory, such as static random access memory, and multi cores of CPU. But the number of cores in CPU is small and each core of CPU has a high price. In this work, we use a popular parallel computing platform, graphic processing unit GPU, to mining core network's super point. We propose a double direction hash functions group which can map hosts randomly and restore them from a dense structure. Because the high randomness and simple process of the double direction hash functions, our algorithm reduce the memory to smaller than one-fourth of other algorithms. Because the small memory requirement of our algorithm, a low cost GPU, only worth 200 dollars, is fast enough to deal with a high speed network such as 750 Gb/s. No other algorithm can cope with such a high bandwidth traffic as accuracy as our algorithm on such a cheap platform. Experiments on the traffic collecting from a core network demonstrate the advantage of our efficient algorithm.