Statistical Determination of Hybrid Threshold Parameters for Entity State Update Mechanisms in Distributed Interactive Applications
Marshall, Damien and McLoone, Seamus and Delaney, Declan and Ward, Tomas (2006) Statistical Determination of Hybrid Threshold Parameters for Entity State Update Mechanisms in Distributed Interactive Applications. In: 10-th IEEE/ACM International Symposium on Distributed Simulation and Real Time Applications, Torremolinos (Malaga), Spain.
Collaboration within a Distributed Interactive Application (DIA) requires that a high level of consistency be maintained between remote hosts. However, this can require large amounts of network resources, which can negatively affect the scalability of the application, and also increase network latency. Predictive models, such as dead reckoning, provide a sufficient level of consistency, whilst reducing network requirements. Dead reckoning traditionally uses a spatial error threshold metric to operate. In previous work, it was shown how the use of the spatial threshold could result in potentially unbounded local absolute inconsistency. To remedy this, a novel time-space threshold was proposed, that placed bounds on local absolute inconsistency. However, use of the time-space threshold could result in unacceptably large spatial inconsistency. A hybrid approach that combined both error threshold measures was then shown to place bounds on both levels of inconsistency. However, choosing suitable threshold values for use within the hybrid scheme has been problematic, as no direct comparisons can be made between the two threshold metrics. In this paper, a novel comparison scheme is proposed. Under this approach, an error threshold look-up table is generated, based on entity speed and equivalent inconsistency measures. Using this look-up table, it is shown how the performance of comparable thresholds is equal on average, from the point of view of network packet generation. These error thresholds are then employed in a hybrid threshold scheme, which is shown to improve overall consistency in comparison to the previous solution of simply using numerically equal threshold values
Repository Staff Only: item control page