Measurement And Improvement of Quality-of-Experience For Online Video Streaming Services

Abstract

HTTP based online video streaming services have been consistently dominating the online traffic for the past few years. Measuring and improving the performance of these services is an important challenge. Traditional Quality-of-Service (QoS) metrics such as packet loss, jitter and delay which were used for networked services are not easily understood by the users. Instead, Quality-of-Experience (QoE) metrics which capture the overall satisfaction are more suitable for measuring the quality as perceived by the users. However, these QoE metrics have not yet been standardized and their measurement and improvement poses unique challenges. In this work we first present a comprehensive survey of the different set of QoE metrics and the measurement methodologies suitable for HTTP based online video streaming services. We then present our active QoE measurement tool Pytomo that measures the QoE of YouTube videos. A case study on the measurement of QoE of YouTube videos when accessed by residential users from three different Internet Service Providers (ISP) in a metropolitan area is discussed. This is the first work that has collected QoE data from actual residential users using active measurements for YouTube videos. Based on these measurements we were able to study and compare the QoE of YouTube videos across multiple ISPs. We also were able to correlate the QoE observed with the server clusters used for the different users. Based on this correlation we were able to identify the server clusters that were experiencing diminished QoE. DynamicAdaptive Streaming overHTTP (DASH) is an HTTP based video streaming that enables the video players to adapt the video quality based on the network conditions. We next present a rate adaptation algorithm that improves the QoE of DASH video streaming services that selects the most optimum video quality. With DASH the video server hosts multiple representation of the same video and each representation is divided into small segments of constant playback duration. The DASH player downloads the appropriate representation based on the network conditions, thus, adapting the video quality to match the conditions. Currently deployed Adaptive Bitrate (ABR) algorithms use throughput and buffer occupancy to predict segment fetch times. These algorithms assume that the segments are of equal size. However, due to the encoding schemes employed this assumption does not hold. In order to overcome these limitations, we propose a novel Segment Aware Rate Adaptation algorithm (SARA) that leverages the knowledge of the segment size variations to improve the prediction of segment fetch times. Using an emulated player in a geographically distributed virtual network setup, we compare the performance of SARA with existing ABR algorithms. We demonstrate that SARA helps to improve the QoE of the DASH video streaming with improved convergence time, better bitrate switching performance and better video quality. We also show that unlike the existing adaptation schemes, SARA provides a consistent QoE irrespective of the segment size distributions.