Dynamically adaptive partition-based interest management in distributed simulation

Abstract

Performance and scalability of distributed simulations depends primarily on the effectiveness of the employed interest management (IM) schema that aims at reducing the overall computational and messaging effort on the shared data to a necessary minimum. Existing IM approaches, which are based on variations or combinations of two principle data distribution techniques, namely region-based and grid-based techniques, perform poorly if the simulation develops an overloaded host. In order to facilitate distributing the processing load from overloaded areas of the shared data to less loaded hosts, the partition-based technique is introduced that allows for variable-size partitioning the shared data. Based on this data distribution technique, an IM approach is sketched that is dynamically adaptive to access latencies of simulation objects on the shared data as well as to the physical location of the objects. Since this re-distribution is decided depending on the messaging effort of the simulation objects for updating data partitions, any load balanced constellation has the additional advantage to be of minimal overall messaging effort. Hence, the IM schema dynamically resolves messaging overloading as well as overloading of hosts with simulation objects and therefore facilitates dynamic system scalability.