Machine Learning-Enhanced Load Balancing for Quantum Chemistry Simulations in Cloud Computing Environments- A Hybrid Approach

Abstract

Quantum chemistry (QC) simulations create a heavy strain on the computation capabilities and are also characterized by many dynamically changing dependant workloads which make the problem of load balancing in cloud environments quite substantial. While many load-balancing techniques generally suffice for a myriad of high-performance computing applications, the same cannot be said for QC tasks. In this paper, a hybrid load-balancing approach that integrates reinforcement learning and predictive modeling is presented to optimize the allocation of resources for QC simulations on the cloud. Specifically, RL is employed for systems resource management while predictive modeling is employed to predict workloads and thus limit latency and optimize usage of cloud resources. Results of experiments show that the use of ML helped the framework to accomplish objectives most possible load-distribution efficiency, categorical improvement of tts, efficiency of up to 20% in load-distribution and reduction of task-completion times. It contributes to the improvement of the efficiency of the QC simulation performed on the cloud infrastructure, while creating preconditions for the further possibility of the application of ML in scientific computing.