High Speed Network Super Points Detection Based on Sliding Time Window by GPU

Abstract

Super points are special nodes in network that contact with lots of other hosts. They are essential in traffic monitoring, measurement, intrusion detection and so on. How to find them in real time is a key step in their applications. All of recent researches focus on mining super points in a fixed time window. They have to wait until the end of a time period to output super points list. This paper proposes a novel algorithm to detect super hosts based on sliding time window. This algorithm can report super points immediately when time window slides forward. GPU has plentiful processor units and abundant graphic memory. To support high speed networking, we utilize common GPU when handling network traffic. According to our experiments, the proposed scheme can deal with core network traffic at throughput as high as 260 million packets per second while the fastest fixed time window algorithm running on CPU could only parse 2 million packets per second. To the best of our knowledge, this is the first super point detection algorithm based on sliding time window and the packet processing speed has not ever been achieved by any other algorithms based on fixed time window.