How We Spend Our Time Online: Predicting Network Traffic Using System Identification. Issue 1 (July 2017)

• Record Type: Journal Article
• Authors: Bhujwalla, Yusuf; Grandemange, Quentin; Gilson, Marion; Laurain, Vincent; Gnaedinger, Eric
• Is Part Of: IFAC-PapersOnLine. Volume 50:Issue 1(2017)
• Page Start: 14125
• Page End: 14130
• Publication Date: 2017-07
• Subjects: Traffic characterisation -- autonomous systems -- nonlinear system identification -- time-series analysis; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2017.08.1854
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: Over the past twenty years, exponential growth of the internet has led to a continuous struggle for content providers to maintain and improve their quality of service. Furthermore, the evolution of network architecture and increased inter-connectivity within the network has changed how traffic is communicated - and how we understand the internet as a whole. Consequently, based on previous work by the authors, this paper formulates an Autonomous System (AS) level approach to traffic characterisation. Such an approach is advantageous given the current network topology, as traffic is overwhelmingly dominated by several agents - with each AS typically exhibiting its own unique, identifiable behaviour. Furthermore, two distinct modelling paradigms are proposed as ways of analysing and predicting network traffic data -from the time-series analysis and system identification communities respectively. The applicability of the proposed modelling techniques is evaluated against real AS-level traffic data, obtained from the Tier-2 European Post Luxembourg network. Results show the suitability of the proposed approaches to the problem, and the simplicity of an AS-level approach from an analysis perspective.