Abstract: The technical advancements of Internet, mobile computing and Internet of things (IoT) have been pushing the deep integration of human, machine and things, which fostered a lot of end-users oriented network search, online social networks, economical business, video surveillance and intelligent assistant tools, which are typically referred to as online data-intensive applications. These new applications are of large scale and sensitive to the service quality, requiring stringent latency performance. However, end-user requests traverse heterogeneous environments including edge network, wide-area network and the data center, which naturally incurs a long-tail latency issue that significantly degrades users’ experience quality. This paper surveys architectural characteristics of edge-computing applications, analyzes causes of the long-tail latency issue, categorizes key theories and methods of the network latency measurement, summarizes long-tail latency optimization techniques, and finally proposes the idea of constructing an online optimization runtime environment and discusses some open challenges.