Novel Approach to Optimize Heterogenous High-Performance Computing Resource Utilization for Hybrid Seismic Imaging Algorithms

A novel optimization solution is presented for managing resources in heterogeneous High-Performance Computing (HPC) environments supporting hybrid seismic imaging algorithms. Existing approaches encounter challenges in achieving efficient resource allocation among diverse applications with disparate computational and memory requirements. Historical job data is analyzed and a formula-based methodology is developed to determine optimal resource allocation. Two key metrics, Application CPU Ratio (ACPR) and Size CPU Ratio (SCR), facilitate accurate identification of applications with high resource demands and inform scheduling decisions.

The results demonstrate a significant reduction in job wait times (>100X decrease) and increased throughput, highlighting the efficacy of the optimization strategy. By mitigating over-provisioning and enhancing resource utilization, the solution yields reductions in operational costs and improvements in scalability. Initially developed for seismic imaging applications, this approach exhibits broad applicability to various HPC environments, offering a robust framework for optimizing resource allocation across diverse computational tasks. This research contributes a systematic approach to improving resource management, yielding enhancements in performance, reduced wait times, and cost savings in HPC systems.