Synchronization Methods for Distributed Agent Based Models

Christine Harvey


Distributed computing addresses the traditional scaling limitations of Agent Based Models (ABMs) and allows for the development of massive-scale models. Synchronization of agent states between multiple processors is complex and needs to be reliable and efficient. The protocol used to synchronize agent states across processors has a significant impact on the efficiency of the tool. Repast HPC is one of the current tools available for distributed ABMs [1]. It approaches the problem by performing a complete synchronization of all entities of interest at every time step. This poster reviews an alternative approach to entity synchronization, a design which manages persistent, pertinent information. This protocol performs an initial synchronization among all entities with relationships to other agents and then only performs updates and synchronization following changes to relevant information. The pertinent data synchronization technique is an event-driven method to manage the communication and synchronization between the processors. The conservative and the event-driven approaches are both described and analyzed in this poster.