Adaptive Container Orchestration Mechanism on Electric Power Supercomputing Clouds
Wei Wei, Ce Wang, Rundong Gan, Haibin Su, Jun Liu · 2023
Electric power supercomputing clouds bring huge computing capacity to various scientific, engineering and commercial applications. In this context, container as a flexible, efficient and scalable virtualization technology has become a key means to improve the resource utilization and operational efficiency of electric power supercomputing clouds. However, existing container orchestration algorithms tend to be static and unable to adapt to dynamically changing loads and environments, resulting in significant resource consumption and failure to meet user quality of service. To this end, we propose an adaptive container orchestration mechanism for power supercomputing clouds that captures the periodic pattern of loads through a time-series model, and design a waiting window approach to mitigate the oscillation phenomenon in resource scaling. Experimental results on two application loads show that our container orchestration mechanism achieves the best service performance and the lowest system resource overhead compared to strong baselines.