Performance Enhancement for Flow Retrievals in Interactive Network Traffic Replay
Flow retrieval is a major performance bottleneck in interactive network traffic replay due to large size of flow table and traffic volume. In this paper we present three new mechanisms that can be integrated with the existing techniques to improve the flow retrieval performance in interactive traffic replay. The new mechanisms are summarized as: 1) Bloom Filter is applied in front of the flow table to reduce the cost of new flow identification. 2) A new technique is developed to early identify the elephant flows on the fly which exploits the large correlations between the rate of the entire flow and its early rate. By capturing the “elephants and mice phenomenon of the Internet traffic we found that the computational cost for retrieving the existing flows could be dramatically reduced. The technique could also benefit many traffic related applications. 3) A cache mechanism is employed to reduce the flow retrieval cost when packets are received from a replay link interface. The cache mechanism captures the locality of the replayed traffic and hence significantly improves the flow retrieval performance. Our experimental results with actual traffic traces show that the new mechanisms are able to effectively reduce the computational cost of flow retrievals and the overall performance of interactive traffic replay is significantly enhanced.
interactive traffic replay flow retrieval elephant flows early identification cache mechanism
Weibo Chu Xiaohong Guan Zhongmin Cai Mingxu Chen
SKLMS Lab and MOE KLINNS Lab, Xian Jiaotong University, Xian, China SKLMS Lab and MOE KLINNS Lab, Xian Jiaotong University, Xian, China Center for Intelligent and Netwo
国际会议
2011 International Conference on Security Science and Technology(ICSST 2011) (2011年安全科学与技术国际会议)
重庆
英文
36-41
2011-01-21（万方平台首次上网日期，不代表论文的发表时间）