An unobtrusive method for tracking network latency in online games

Abstract

Online games are a very important class of distributed interactive applications. Their success is heavily dependant on the level of consistency that can be maintained between participants communicating in the virtual world. Achieving a high level of consistency usually involves the transmission of a large amount of network traffic. However, if the underlying network connecting participants is unable to process this traffic, then network latency will increase, which will in turn negatively impact on consistency. Many schemes exist which attempt to reduce network traffic, and thus reduce the effect of network latency on the interactive application. However, applications that employ these schemes tend to do so with little knowledge of the underlying network conditions, and assume a worst-case scenario of limited bandwidth. Such an assumption can actually cause these latency reduction schemes to perform sub-optimally, and ironically introduce more inconsistency than they reduce. Hence, it is important that online game applications become aware of network conditions, such as available bandwidth. Existing methods of estimating bandwidth operate by analysing trends in one-way latency, and require that extra data be transmitted between nodes in order to capture the latency trends. Such an approach does not suit online games, as the extra data requirements could increase network latency, and affect the ability of the application to scale to multiple participants. To deal with this issue, this paper proposes a method by which online games can unobtrusively track one-way network latency. This method requires no time-stamping information to be transmitted between participants and operates using data already being transmitted as part of the online game application, meaning that its impact on the network is minimal. NS2 simulations demonstrate that the trends collected by this method can be used to estimate bandwidth under certain conditions.

Optimizing consistency by maximizing bandwidth usage in distributed interactive applications

“…However, this requires extra time-related data to be transmitted within each packet, and is subject to clock skew effects. To deal with this issue, our scheme employs an unobtrusive method of tracking latency trends [Marshall et al 2007]. This method operates by predicting when packets should arrive from a sender, given that the sender is transmitting packets at a constant rate.…”