| Bio | 
| Paper |
Robert Zinchak
Trinity University
Subject Listing - Computer Science
Advisor: Dr. Gerald Pitts
Thursday, Oral Session 3, Presentation 5, Robinson Hall 239
DISTRIBUTED COMPUTATION IN AN INTERACTIVE ENTERTAINMENT ENVIRONMENT
The proliferation of multiplayer games has led to an increase in the total network capacity for processing in games, but this capacity is rarely fully utilized or balanced. One prominent problem of distributed processing in a gaming environment is increased latency time, which causes player disinterest in the game potentially causing poor sales and the termination of future commercial development of this technology. Existing distributed techniques such as OpenMP, MPI or VMPI are not well suited to gaming applications and may introduce additional overhead.
It is the purpose of this paper to describe a simple, yet effective technique (based on existing ideas in both multiplayer network games and parallel processing) that can be implemented into existing networked games to allow for distributed computation, which in turn will balance the work load across all connected machines while simultaneously decreasing visible latency to players. Most multiplayer games use a client-server architecture, where the server acts as a referee between clients to manage the game state. A game world with many A.I. agents (simple non-playing characters) overloads the server and causes higher latency levels. Load may be balanced by allowing trusted clients to compute these actions. Development and maintenance costs can be kept low while not diminishing production quality, leading to a successful game. In addition to gaming applications, this technique can be incorporated in any existing client-server network application, including: communication, transportation, and process control.
This paper seeks to discover whether parallel processing offers a viable option for reducing latency and enabling larger, more realistic virtual worlds in multiplayer games. A comparison between the traditional client-server method and the distributed processing approach will be provided. An example is created that simulates many A.I. agents in a larger N.I. (narrative intelligence) environment. Empirical tables and graphs will show the relationship between overhead and optimization this technique creates, illustrating whether parallel processing in networked multiplayer games deserves implantation.
Advisor: Dr. Gerald Pitts, Caruth Distinguished Professor, Computer Science, Trinity University, San Antonio, TX