Section: New Results

Service Transparency

On active sampling of controlled experiments for QoE modeling

Participants: Muhammad Jawad Khokhar, Nawfal Abbasi Saber, Thierry Spetebroot, Chadi Barakat.

For internet applications, measuring, modeling and predicting the quality experienced by end users as a function of network conditions is challenging. A common approach for building application specific Quality of Experience (QoE) models is to rely on controlled experimentation. For accurate QoE modeling, this approach can result in a large number of experiments to carry out because of the multiplicity of the network features, their large span (e.g., band-width, delay) and the time needed to setup the experiments themselves. However, most often, the space of network features in which experimentations are carried out shows a high degree of uniformity in the training labels of QoE. This uniformity, difficult to predict beforehand, amplifies the training cost with little or no improvement in QoE modeling accuracy. So, in this work, we aim to exploit this uniformity, and propose a methodology based on active learning, to sample the experimental space intelligently, so that the training cost of experimentation is reduced. We prove the feasibility of our methodology by validating it over a particular case of YouTube streaming, where QoE is modeled both in terms of interruptions and stalling duration. This first validation has appeared in [19]. In another paper which is currently under submission, we propose an online version of this methodology together with a set of criterion to stop the experiments when the learner is confident enough.

On the Cost of Measuring Traffic in a Virtualized Environment

Participants: Karyna Gogunska, Chadi Barakat, Guillaume Urvoy-Keller, Dino Lopez Pacheco.

The current trend in application development and deployment is to package applications and services within containers or virtual machines. This results in a blend of virtual and physical resources with complex interconnection network schemas mixing virtual and physical switches along with specific protocols to build virtual networks spanning over several servers. While the complexity of this set-up is hidden by private/public cloud management solutions, e.g. OpenStack, this constitutes a challenge when it comes to monitor and debug performance related issues. In this work, carried out in collaboration with the Signet team of I3S with the support of the UCN@SOPHIA Labex, we introduce the problem of measuring traffic in a virtualized environment and focus on one typical scenario, namely virtual servers interconnected with a virtual switch. For this scenario, we assess the cost of continuously measuring the network traffic activity of the machines. Specifically, we seek to estimate the competition that exists to access the physical resources (CPU, memory, etc.) of the physical substrate between the measurement task and the legacy application activity. The results of this first study are currently under submission.

LISP measurements

Participant: Damien Saucez.

The Locator/Identifier Separation Protocol (LISP) separates classical IP addresses into two categories: one for identifying terminals, the other for routing. To associate identifiers and locators LISP needs a specific mechanism, called mapping system. This technology is still at an early stage but two experimental platforms have already been deployed in the Internet: LISP Beta Network and LISP-Lab. However, only the LISP Beta Network is monitored with LISPmon that partially monitors the mapping system once a day. To accompany the growth of LISP, a dynamic and complete monitoring system is required. Therefore, we propose LISP-Views, a dynamic versatile large scale LISP monitoring architecture. LISP-Views allows to automatically conduct comprehensive and objective measurements. After running LISP-Views in the wild for several months and comparing the monitoring results with LISPmon, we confirm that LISP-Views provides more detailed and accurate information. We observe the different behaviours between every network entity within mapping system, and also explore the current LISP performance for further improvements. A paper on "LISP-Views Monitoring LISP at Large Scale" was published in ITC this year.