Relay strategy in online mobile games: a data-driven approach

Abstract

With the booming of online mobile games (OMGs), game operators need to provide high-quality game service for users. Using relay has become the de factor approach for game streaming today, because it is easy to use (e.g., game sessions can be redirected via CDN servers) and has good scalability. Today, it has become the norm rather than the exception for game operators to hire CDN servers for their game services in a pay-per-use manner to serve massive users. Given the limited resource, selecting game sessions which are relayed has become a critical decision that can significantly affect users' quality of experience (QoE). Conventional strategies are generally rule-based, e.g., assigning game sessions to relay paths according to their past network performance, but cannot guarantee any particular QoE level because network performance dynamically changes. In this paper, we propose using data-driven approach to study network performance of game sessions in temporal and spatial patterns. Our findings indicate that there is obvious regularity for network performance of game sessions in temporal and spatial patterns. We design a machine learning-based predictive model to capture the quality of a game session given particular network performance metrics. Based on that, we strategically assign game sessions to relay paths to maximize the overall QoE. Trace-driven experiments are used to demonstrate the effectiveness and efficiency of our design.