Research On A Hierarchy-based Workload Dividing Approach For The Distributed DDM

The primary goal of data distribution management (DDM) in High LevelArchitecture (HLA) is to reduce the amount of data received by federates during adistributed simulation. In recent years, many distributed DDM approaches have beenproposed to distribute the DDM processing workload among hosts. However, the existingdistributed DDM approaches do not consider the situation that the distribution of regionsin the routing space is in a relatively static state during the simulation, i.e. a simulationobject is likely to publish/subscribe regions in a certain area during a period of time. Inorder to improve the performance of distributed DDM in this situation, a hierarchy-basedDDM problem and its formal definition are presented.The workload division method plays a key role in the performance of distributedDDM approaches. Considering advantages and disadvantages of existing approaches, thispaper proposes a hierarchy-based workload division approach to efficiently distribute theworkload among hosts and to reduce the notification messages which the hosts receive aswell as the delivery time of region update messages. The approach consists of two stages:(1) static division applied at the simulation initialization;(2) dynamic division used duringthe simulation. In the initial stage, the routing space is divided into several blocks basedon the initial region distribution. Then these blocks are distributed among host groups, andeach block is further divided into subdivisions in each host group. After the static division,each host group works relatively independently and dynamically redivides its own blockinto new subdivisions to keep workload balanced. As the simulation progresses, therelative independence among groups may be compromised. Therefore, a controlmechanism is introduced at appropriate time to redivide the routing space into new blocks.By using the hierarchy strategy and the dynamical re-division of blocks and subdivisions,our algorithms can more effectively handle the relatively static region distribution in thedistributed environment.The experimental results indicate that our algorithm not only can effectively realize data distribution and balance workload among hosts, but also can achieve betterperformance than existing approaches in the real running environment.