Evolutionary Multi-Objective Task Scheduling for Heterogeneous Distributed Simulation Platform

Most existing distributed simulation platforms lack native support for Python scripts, thereby hindering the seamless integration of AI models developed in Python. Some simulation platforms support script languages like Lua or javascript, but scheduling tasks in heterogeneous simulation platforms that are composed of simulation engine and script engine is a challenging problem. Moreover, conventional task scheduling methods often overlook the simulation time constraints, which are essential for simulation synchronization. In this paper, we propose a Heterogeneous Distributed Simulation Platform (HDSP) that could integrate different script languages, especially Python, to empower the simulation by leveraging intelligent AI models. A Dynamic Multi-Objective Optimization (D-MO) Scheduler is also designed to efficiently schedule simulation tasks that run across heterogeneous simulation engines and satisfy simulation synchronization constraints. HDSP integrates various script engines, enhancing its adaptability to model dynamic simulation logic using different script languages. D-MO Scheduler optimizes Simulation Acceleration Ratio (SAR), Average Weighted Waiting Time (AWWT), and Resource Utilization (RU). The D-MO scheduling problem is characterized as an NP-hard problem, tackled using the NSGA-III algorithm. The simulation time synchronization constraints are implemented through Lower Bound on Time Stamp (LBTS) and lookahead approach. The comparative results and statistical analysis demonstrate the superior efficacy and distribution performance of proposed D-MO Scheduler. The proposed HDSP and D-MO Scheduler significantly boost the capability to support Python-based AI algorithms, and navigate complex scheduling demands efficiently.