Research on Computational Load Balancing for Massively Parallel Tasks Based on Adaptive Iterative Algorithm

Focusing on the load balancing problem in massively parallel task computing, this study proposes a task scheduling model based on an adaptive iterative algorithm. The model effectively solves the shortcomings of traditional static scheduling algorithms in the face of heterogeneous computing resources by dynamically adjusting the task allocation strategy. In the study, we first design a mathematical model for load-balanced scheduling of massively parallel task computation, and then use the improved adaptive genetic algorithm (AGAPD) to solve for the balanced state. The algorithm combines the global search capability of genetic algorithm and the local optimization capability of adaptive mechanism, and is able to achieve efficient load balancing in complex cloud computing environments. The experimental results show that the AGAPD algorithm saves 90.56%-93.99% of time compared with the traditional algorithm in terms of task completion time at 2000 node size, and its CPU occupancy rate is also reduced by 41.73%-47.27% compared with that of the traditional algorithm, which improves the efficiency of solving this type of problems. In addition, the dynamic load balancing overhead of the model remains around 0.18, and its speedup ratio increases significantly with the increase of system size, which shows that the model can significantly improve the overall performance of massively parallel computing systems.