Multiobjective message scheduling for Hybrid Synchronization in Distributed Simulations

Abstract

One of the essential aspects of distributed simulations is to order events according to a causal consistency model. Traditionally, implementing causal consistency can be made using a conservative or optimistic approach. However, traditional techniques are costly in processing time to ensure causality. A promising approach to order events is a hybrid synchronization approach, where processes can change dynamically between optimistic and conservative approaches. Unfortunately, synchronizing processes running a hybrid synchronization is a complex problem. In this work, we discuss a multi-objective scheduling of hybrid synchronization messages problem. Beyond that, we propose using a scheduling algorithm to reach an equilibrium between processing and causality violations and describe how to integrate the algorithm in an existing distributed simulator. The algorithm uses two memoization phases, making the scheduling suitable for a dynamic environment. Finally, to demonstrate the feasibility of our scheduling approach, we implemented it in an existing distributed simulation architecture. Analysis based on the experiments demonstrates the behavior of the simulation regarding the number of discarding/processed messages and work performed.