An intelligent scheduling algorithm for complex manufacturing system simulation with frequent synchronizations in a cloud environment

For cloud-based, large-scale complex manufacturing system simulation (CMSS), allocating appropriate service instances (virtual machines or nodes) is a promising way to improve execution efficiency. However, the complex interactions among and frequent aperiodic synchronizations of the entities of a CMSS make it challenging to estimate the influence of service instances' computing power and network latency on the execution efficiency. This hinders the appropriate allocation of service instances for CMSS. To solve this problem, we construct a performance estimation model (PEM) using the executed events and synchronization algorithms to evaluate the running time of CMSS on different service instance combinations. Further, an intelligent scheduling algorithm that introduces PEM as fitness function is proposed to search for a near-optimal allocation scheme of CMSS service instances. To be specific, the PEM-based optimization algorithm (PEMOA) incorporates simulated annealing into the mutation phase of a genetic algorithm to strengthen its local searching ability. A series of experiments were performed on a computer cluster to compare the proposed PEMOA with two representative algorithms: an adapted first-come-first-service-based and the max-min-based allocation algorithms. The experimental results demonstrate that the PEMOA can reduce the running time by more than 7%. In particular, the improvement of PEMOA increases when the manufacturing system simulation is communication-intensive or spans a small number of service instance combinations.