Title: Multi-armed bandit algorithms over DASH for multihomed client

Authors: Ali Hodroj; Marc Ibrahim; Yassine Hadjadj-Aoul; Bruno Sericola

Addresses: Faculty of Engineering, Saint Joseph University of Beirut, ESIB, Mar Roukos, Lebanon ' Faculty of Engineering, Saint Joseph University of Beirut, ESIB, Mar Roukos, Lebanon ' Univ Rennes, Inria, CNRS, IRISA, Rennes, 35042, France ' Univ Rennes, Inria, CNRS, IRISA, Rennes, 35042, France

Abstract: Mobile customers are increasingly demanding video traffic, which has accounted for the majority of mobile data traffic over the past two years. To improve the quality of video received by multi-homed clients, a network selection algorithm based on multi-arm bandit heuristics is proposed on top of the most widely used standard for video streaming, dynamic adaptive streaming over HTTP (DASH). For the selection of interfaces, DASH uses the default one and assigns the quality according to the adaptive bit rate algorithm, without examining the conditions of alternative networks that could offer higher quality. Subsequently, few adjustments are required to enhance the video quality. Two algorithms (UCB and Epsilon Greedy) were embraced for improving MPEG-Dash. The investigations are performed through a proving ground execution, which show that UCB surpasses Epsilon Greedy, in stable system conditions, while discovering the best compromise between searching for new choices and overusing the triumphant variation.

Keywords: DASH; dynamic adaptive streaming over HTTP; multi-homed; video streaming; multi-armed bandit; UCB; Epsilon greedy; reinforcement learning; machine learning; QoE.

DOI: 10.1504/IJSNET.2021.119485

International Journal of Sensor Networks, 2021 Vol.37 No.4, pp.244 - 253

Received: 10 Feb 2021
Accepted: 10 Feb 2021
Published online: 07 Dec 2021 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article