Quality of Experience of WebRTC based video communication

Abstract

Online video applications are growing in popularity and using an increasing share of the consumer Internet traffic. Web Real-Time Communication (WebRTC) is a new technology which allows browser-to-browser communications without any software downloads or user registration. The focus of this report is the Quality of Experience (QoE) in the context of WebRTC.

We have created a fully controllable testing environment, a testbed, where we can manipulate a network to perform under various conditions by altering the parameters packet loss rates, Mean Loss Burst Size (MLBS), delay, jitter, Central Processing Unit (CPU), and bandwidth. A testbed is of importance for testing of QoE services in general, and also for application developers because they can analyze their application s behavior in altered networks which can simulate real-world use.

We have used the WebRTC application appear.in for several different experiments where we altered the network conditions. We have col- lected both connection statistics and the subjective feedback from each participant.

Firstly, we conducted a pilot study consisting of two-party conversa- tions of 12 participants, where our main focus was on packet loss and MLBS. After that, we conducted three-party conversations where we tested packet loss, MLBS, delay, jitter, and CPU.

We found in our experiments that the perceived quality of a specific packet loss rate depends also on the MLBS. Higher MLBS seems to result in an overall worse user experience, especially impacting the audio quality of the conversation. We also found that delay (<1 second) does not necessarily leads to a worse user experience, while jitter quickly impacts both audio and video quality. Finally, it seems that the CPU limitations seem to affect only the user with the reduced CPU-usage.

The experiments show that the testbed is working as specified, and can be used for more extensive research in the future.

Keywords - WebRTC, Quality of Experience, appear.in, testbed, pilot study, Mean Loss Burst Size.