A binary partition-based matching algorithm for large scale distributed simulation

Abstract

Data distribution management (DDM) is one of the most important filtering mechanisms in large scale distributed simulations. DDM has been successful in reducing the network traffic in some respect, but its capability is limited by the computational overhead for matching update regions and subscription regions that represent the interests of data producers and data consumers, respectively. For example, a typical DDM scenario is a battlefield simulation, where sensors are deployed to detect enemy movement. The enemy units are moving to a certain destination point and a sensor detects any enemy units. The update regions represent the location of enemy units with small rectangles, and the subscription regions represent the detecting ranges of the sensors. In this example, only if the update regions of the enemy units and subscription regions of the sensor overlap, the location of the enemy units will be routed by the DDM filtering mechanism. However, that mechanism produces high computational overhead because a matching process is performed to calculate the intersection between all pairs of update regions and subscription regions. As the previous matching algorithms do not consider the characteristics of the region distribution, it is difficult to select a matching algorithm that is appropriate for some region-distribution situations. In the case of large scale distributed simulations, such as a battlefield simulation, it will probably be a situation with large overlapping regions. The computational overhead of the matching process in large scale simulations is significant, because of superfluous intersections between regions.

First, in this dissertation, we propose a binary partition-based matching algorithm based on the divide-and-conquer approach to reduce the computational overhead in the matching process. The design goal of our algorithm is to perform binary partitioning that divides the regions into two partitions, the left-hand partition and right-hand pa...