Time Management for Virtual Worlds based on Constrained Communication Model Umar Farooq and John Glauert School of Computing Sciences, University of East Anglia, Norwich NR4 7TJ, United Kingdom. (u.farooq, j.glauert)@uea.ac.uk Abstract—Time Management (TM) is an integral part of the parallel and distributed systems that maintains the temporal order of events in a system. In this paper, we present a decentralised TM approach using a constrained communication model based on the inherent properties of virtual worlds. The proposed method uses a flat communication model and a region synchronises itself with a set of regions that share boundaries with it. It is evaluated with the help of a simple simulation model and compared with non-synchronised and decentralised scenarios. The simulation results show that it maintains local causality constraint and reduces communication of messages over the network. It is potentially more scalable and minimises longer delays and complexity compared with hierarchical strategies with multiple levels. I. I NTRODUCTION The horizon of Parallel and Distributed Simulations (PADSs) has recently been extended for a number of different applications. However, Virtual Environments (VEs) among these applications have gained much attention from the end users. The reasons include the provision of impressive online content, a significant increase in speed, and a significant decrease in prices of high speed computers and network resources. Most recently, people are showing a great inter- est in social collaborative spaces called virtual worlds. The users have freedom in content development according to their desires on virtual land that is purchased using inland virtual currency [1][2][3]. Since, a PADS system is partitioned and executed with the help of a set of dedicated computers, it is a challenging task to maintain temporal order of events. This issue is called Time Management (TM) (alternatively synchronisation or consistency) and it is an integral part of a PADS system. Parallel and distributed systems are fault tolerant, scalable and provide better interactive user experience but their performance is degraded when conservative approaches are used for time synchronisation [4]. The existing algorithms for PADS systems have shown great success for their target applications but they have major performance issues when used for virtual worlds. The target application of this research is a world like Second Life [2][3] that imitates the nature of social activities of the physical world. It gives more freedom than physical world and has an inherent property that a user or an activity is only influenced by the users or activities surrounding them. Based on this property, we put restrictions on communication and data distribution, and a system (serving a region) synchronises itself with the regions (defined as adjacent regions) that share physical boundaries with it. This work proposes an efficient decentralised TM approach based on the inherent properties of a virtual world. It is based on a conservative event driven approach and maintains a consistent view of dynamic hierarchical models based on our split strategies [5]. Control is fully decentralised and a server takes purely local decisions in consultation with adjacent servers. It maintains traditional constraints and guarantees that all events are processed in their temporal order. It greatly reduces communication overhead, complexity, and delays by avoiding a number of intermediate points compared with traditional mechanisms based on conservative approaches. This paper is organized as follows. Section II gives the background of this research. The proposed decentralised TM scheme is presented in section III. Section IV provides the simulation results and illustrations. Conclusions and future work are presented in section V. II. BACKGROUND AND MOTIVATION According to Fujimoto [6], the basic goal of TM is to get exactly the same results as a sequential computer while executing a system with a set of servers. Each system main- tains a list of events (both internal and external) and in each iteration removes the smallest timestamped event from the list and processes it in correct temporal order (the local causality constraint). Early synchronisation attempts proposed by Bryant [7], and Chandy et al. [8] are prone to deadlock. To resolve this issue, the concept of null messaging is used to advance the simulation clock of a process. It uses a value known as the Lookahead constraint that determines a safe range for event processing. The Lookahead value has a dra- matic performance effect on the TM algorithm [9]. According to Pan et al. [10], asynchronous TM algorithms with small Lookahead values have a “time creep” problem. The main drawback of the null message algorithm is that it generates an excessive number of null messages thus introducing longer