Network Traffic Anonymization Based On Multiview Approach

Network traffic anonymization is a classical technique that is used to protect privacy before the traffic outsourced by the enterprise.To obtain professional analysis results,enterprises have an increasing demand for outsourcing network traffic to thirdparty analysts.However,sensitive information in the traffic,such as IP addresses,exposes privacy risk,which needs to be anonymized before outsourcing.Preserving the privacy and utility of network traffic is a critical problem of all time.This dissertation focuses on the research of anonymization that preserves both privacy and utility.The multi-view approach has received extensive attention on the problem of preserving both privacy and utility of anonymized network traffic.Traditional network traffic anonymization approaches have a tradeoff between privacy and utility,either greatly damaging the useful information in the traffic,or being vulnerable against malicious attacks.The multi-view approach preserves both utility and privacy by shifting this tradeoff from privacy and utility to privacy and computational cost.The key idea is to generate multiple pseudo-copies of network traffic and outsource them together with the real one.Current multi-view approach based on pseudo random number generator(PRNGMV)preserves both privacy and utility,but it has drawbacks.Firstly,PRNG-MV cannot ensure the indistinguishability between the real traffic and pseudo ones,combining which attack could exists to infer the real one.Meanwhile,traffic anonymized by PRNG-MV encounters a utility loss.Aiming at above problems,this dissertation conducts an in-depth study on the privacy risk and countermeasures,as well as the utility requirements of multi-view approach.The main research contents of this dissertation are as follows:(1)This dissertation proposes a real view inference attack against multi-view approach.Based on the entropy of different IP prefix partitions,this attack could be exerted to identify the real network traffic from multiple network traffic.(2)This dissertation analyzes the requirements of preserving both privacy and utility combining real view inference attack and proposes a new multi-view approach based on the IP prefix circle and IP distribution(PI-MV)that could satisfy the requirements.(3)On the basis of above research on attack and defense,this dissertation designs and implements a tool package for multi-view network traffic anonymization.In addition,a secure IOT traffic outsourcing(SITO)system is implemented for deployment under the IOT environment.The comparative analysis of the evaluation results shows that real view inference attack could infer the real traffic of PRNG-MV at a high ratio and PI-MV provides a guarantee on privacy,utility,and computational cost.Also,the deployment experiment shows that the SITO system deployed in an edge router has a high reliability on collecting and anonymizing network traffic.