An autonomous QoE-driven network management framework Janne Seppänen ⇑ , Martín Varela, Aggeliki Sgora VTT Technical Research Centre of Finland, PL 1100, Oulu 90571, Finland article info Article history: Available online 4 December 2013 Keywords: Quality of experience Multimedia Over-the-top Management system Access point Admission Control QoS Monitoring abstract Recently, network researchers have taken a great interest in quality of experience (QoE) and in the new aspects it brings in the study of the link between network conditions and user satisfaction. Also, the real- ization that the information of users’ satisfaction can be directly applied in the network management in a real-time manner has resulted in a fair amount of publications. Although the systems and frameworks presented in these publications tackle the subject of QoE-driven management quite successfully, they often concentrate on certain applications or technologies. We present a generic QoE management frame- work, which is applicable to a broad range of systems. We also demonstrate an instantiation of this framework as a network access point management system for RTP-based video. This system is not only able to positively affect the perceived quality of the multimedia application considered, but also to reduce over-prioritization and optimize resource usage. Ó 2013 Elsevier Inc. All rights reserved. 1. Introduction Multimedia services, and in particular video, make up a large and ever-increasing portion of the total Internet traffic. The popu- larity of the so-called ‘‘Over-the-Top’’ (OTT) services, such as You- Tube, Netflix, Hulu, and other Web-based video services has exploded and will continue to do so (e.g. with the adoption of Web- RTC [1] for real-time browser-based communications), as has their importance to users and business. This fact, coupled with the avail- ability of fast and cheap mobile connectivity and mobile devices capable of displaying high-definition content, poses a serious chal- lenge to mobile operators, as users become accustomed to more re- source-demanding services and demand better quality. In contrast to operator-run media services (such as IPTV, mobile TV, or IMS- based ones), where the operator has control over the whole chain of transmission, OTT services come from outside the operator’s network, and the operators have very little, if any, control over them (some content providers such as Netflix work with operators to provide caching servers within the operator’s network, but this is not generally the case with all content providers). The operator’s challenges are thus many; firstly, it has to deal with the increasing demand on their infrastructure, notably so over the last hop (base stations and WiFi hotspots, for example). Sec- ondly, OTT video is most commonly delivered over HTTP, which makes it hard to separate from other web traffic. Finally, from the operator’s point of view, OTT services are hard to monetize (the content providers get the revenue and the operator just sees an increased use of resources) and at the same time, if the quality of these services is not good enough for the users, the operator will face higher user churn. In this paper we address these challenges by providing traffic control mechanisms based on a combination of Quality of Experience (QoE) 1 estimations, and subscriber and application- based traffic differentiation. We present a framework to instrument QoE-driven network management mechanisms, and in the context of this framework, we implement a concrete prototype for QoE-driven control. We expand upon our previous work [2] by incorporating network performance models, allowing the proposed approach to always make the right decision by predicting the possible outcomes, instead of just reacting to a drop in quality and hoping that the reac- tion will result in a positive change. The goal of the proposed work is to allow operators to properly address the needs of their users (in terms of QoE), while introducing subscriber differentiation as a means of increasing revenues and simplifying resource allocation (i.e. customers who pay more are prioritary). The proposed approach is able to (a) identify the relevant media flows, (b) estimate their cur- rent QoE, (c) select the appropriate priority for the flows based on their application type, subscriber class, current QoE for it and other media flows, and expected QoE after the control mechanism kicks in (based on network performance models) and (d) perform access con- trol on new flows based on the current quality for existing flows, and the incoming flow’s application and subscriber class. The rest of the paper is organized as follows. Section 2 provides an overview of related works. Section 3 describes the proposed traffic management system. Sections 4 and 5 present a prototype 1047-3203/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jvcir.2013.11.010 ⇑ Corresponding author. E-mail addresses: janne.seppanen@vtt.fi (J. Seppänen), martin.varela@vtt.fi (M. Varela), ext-angeliki.sgora@vtt.fi (A. Sgora). 1 More specifically, perceived quality. J. Vis. Commun. Image R. 25 (2014) 565–577 Contents lists available at ScienceDirect J. Vis. Commun. Image R. journal homepage: www.elsevier.com/locate/jvci