An Empirical Study of the WeChat Mobile Instant Messaging Service

Mobile messaging applications are already a part of everyone's daily life. Despite their prevalence, we have limited knowledge on user behavior and data transmission performance of these services. In this paper, we examine WeChat, one of the largest mobile messaging services with over 800 million active users. To this end, we analyze a packet-level dataset captured in a cellular network, containing 121GB WeChat data and 4.8M WeChat flows. Our analysis reveals several unique features of user behavior in WeChat, including the diurnal traffic pattern with burst spikes, dominated traffic by media flows, burst server-to-client messages, among others. Perhaps more importantly, we leverage off a machine learning algorithm to classify users into several clusters, of which each captures one typical usage pattern. Besides, we find a non-negligible portion of media flows failed to completely transmit the media objects, mostly due to network-related factors. One of such factors is the inefficiency of TCP to recover from packet loss at the end of transmission.