Towards end-to-end sla assessment

Abstract

The introduction of a new set of services to the Internet impose a series of tight conditions on the network usage: there is need of more bandwidth, bounded packet losses or limited delays on data delivery. Until now, the only considered parameter was the bandwidth. Nevertheless, this paradigm is changing because of the more demanding services being deployed, namely VoIP, Videoconferencing or Video-Streaming. The above requirements are easily met on reduced environments or small networks, where the resources are managed by a single administrative unit, and Quality of Service (QoS) mechanisms are not difficult to deploy. Nevertheless, the problem has a much more difficult solution when dealing with Inter-Domain environments. There, the traffic policies, the network usage, and the resources in general are managed by different administrative units, which most likely will have different goals and policies when distributing the available resources. On top of this, every day more and more customers are requiring guaranties for their services, and operators want to know the network status all the time. Therefore, there is an increasing need of monitoring whether the network can provide the desired levels of quality. Or, what is the same, if the network is complying with the contracted Service Level Agreements (SLA). This thesis has the major goal of developing an on-line distributed SLAAssessment infrastructure, which will give live reporting about the quality the network is providing. We focus on the Intra-Domain study, and outline several alternatives for extending this to the Inter-Domain area. To develop such infrastructure, some challenges have to be addressed. First there is the need of specifying the techniques to be used in order to monitor and analyse the traffic in real-time. Second, there is the study of the resources in terms of bandwidth required to perform the assessment. And third, there is a trade-off between the accuracy and the resources to be decided. We solve the problem of the resource requirements by proposing a set of innovative sampling techniques, which make an efficient use of the resources. Specifically, the main contributions of this work are: We develop a Network Parameter Acquisition System (NPAS), that is a scalable distributed SLA Assessment infrastructure that monitors the QoS of the sensible flows on the network. We analyse the current measurement techniques, proposing ways of using the knowledge acquired from metrics behaviour and Internet in general to extend and optimise NPAS. We optimise NPAS by using static and dynamic adaptive sampling techniques in order to control the resources needed by the system. In parallell we improve classical Quality of Experience (QoE) algorithms in order to work with current network technologies and adapt them to our NPAS, delivering a system which can assess the user perceived quality of specific flows. Finally we propose a novel way of decoupling the metric computation from the quality of the network by smart distance algorithms over Inter- Packet Arrival Times. With this decoupling we can assess the QoS of a network with minimum metric computation, with the consequent boost in performance of the system.