Netlag: a performance evaluation tool for massively multi-user networked applications

Large-scale Massively Multiplayer Online Games (MMOGs) and other networked applications pose challenging performance requirements such as low response times (down to 100 ms for action games like First-Person Shooters) and high update rates (up to 50 Hz). They require multi-server architectures in order to scale to higher player numbers (up to 10⁵ in a single application session). During the application development process, it is necessary to study performance properties such as response time to user actions or CPU consumption in order to optimize the application. To analyse performance properties, the application developer needs to (i) model these properties, (ii) collect information about the variables of interest, and (iii) process the collected information to study the results. In this paper, we propose a novel tool set (Netlag) that supports the collection and processing of variables of interest in the context of MMOGs. The tool set consists of a C++ library that allows to collect and store information in a generic way, as well as a Java application that visualises the collected information. As a case study, we conduct how Netlag is used to evaluate the performance and scalability of an example MMOG application. Furthermore, we describe measurements of the overhead introduced by Netlag which demonstrate that its application intrusion is minimal, thus proving its good applicability for MMOGs and other networked applications.