Section: New Results

Machine Learning for an efficient and dynamic management of network resources and services

Machine Learning in Networks

Participants : Dana Marinca, Nesrine Ben Hassine, Pascale Minet.

Machine learning techniques can be used to improve the quality of experience for the end users of Content Delivery Networks (CDNs). In a CDN, the most popular video contents are cached near the end-users in order to minimize the contents delivery latency. The idea developed hereafter consists in using prediction techniques to evaluate the future popularity of video contents in order to decide which ones should be cached. The popularity of a video content is evaluated by the number of daily requests for this content.

We consider various prediction methods, called experts, coming from different fields (e.g. statistics, control theory). To evaluate the accuracy of the experts' popularity predictions, we assess these experts according to three criteria: cumulated loss, maximum instantaneous loss and best ranking. The loss function expresses the discrepancy between the prediction value and the real number of requests. We use real traces extracted from YouTube to compare different prediction methods and determine the best tuning of their parameters. The goal is to find the best trade-off between complexity and accuracy of the prediction methods used.

We also show the importance of a decision maker, called forecaster, that predicts the popularity based on the predictions of selection of several experts. The forecaster based on the best K experts outperforms in terms of cumulated loss the individual experts' predictions and those of the forecaster based on only one expert, even if this expert varies over time. This study has been presented at the IWCMC 2016 conference [18].

In the paper presented at the WiMob 2016 conference [18], we apply these prediction methods to caching. We first selected the best experts in charge of predicting the popularity of video contents in real traces of YouTube. We tuned the parameters of the DES expert. We proved that the well-known LFU caching strategy can also be considered as a prediction based strategy on the Basic expert. Simulation results show that the DES prediction-based caching strategy provides similar Hit Ratio to the well-known LFU caching strategy. These results are usually close to the optimal ones that can be achieved only when knowing in advance the popularity of each video content for the next day, which is an unrealistic assumption. The exception occurs when a content whose popularity was very poor becomes suddenly very popular with millions of solicitations. In such a case, the accuracy of prediction methods becomes poor. This opens a research direction where the knowledge of societal events is taken into account to improve the prediction.