Traces for network applications provide insights of application-specific traffic patterns and network requirements among other things. One imminent use of trace files, given the trace files closely resemble the behavior of real life users, is building the traffic model of virtual worlds. Several research groups have pointed out that modeling network traffic of virtual worlds is valuable for: (i) ISPs to configure their networks, (ii) virtual world software developers to achieve better QoS, and (iii) research communities to generate synthetic traffic for simulations. A few approaches can be applied to refine the existing models in the literature. For example, packets in each channel, or packets of the same type can be modeled with a separate distribution. In addition, region characteristics, such as avatar/object density, can be considered as parameters to traffic models. These extensions, however, are not possible without our extensive network traces collected from diverse environments.
Network traces also have the potential to improve user experience. Previous research has observed that some packet types are very sensitive to packet loss and delivery delay, while others are more loss and delay tolerant. This leads to a possible QoS mechanism that assigns packets to different DiffServ classes based on their packet types for better user experience. Other QoS mechanisms are also possible. For example, modern smartphones are equipped with multiple network interfaces, and viewers on smartphones can send different types of packets via different interfaces to prolong battery life, which is critical to mobile devices. This can be done because although WiFi networks generally achieve higher bandwidth than 3G networks, WiFi networks consume more energy when the traffic volume is low. Designing and evaluating these QoS mechanisms heavily rely on our network traces and the traffic models derived from them.
Although network traces of virtual worlds are valuable to ISPs, virtual world software developers, and research communities, they do not exist in the public domain. In this project, we implement a complete testbed to efficiently collect and analyze network traces from a popular virtual world: Second Life. We use the testbed to gather traces from 100 regions with diverse characteristics. The network traces represent more than 60 hours of virtual world traffic, and we create trace files in a well-structured and concise format. Our preliminary analysis on the collected traces is consistent with previous work in the literature. It also reveals some new insights: for example, local avatar/object density imposes clear implications on traffic patterns. The proposed testbed and released trace files can be leveraged by research communities for various studies on virtual worlds. For example, accurate traffic models can be derived from our trace files, which in turn can guide developers for better virtual world designs.