3PEIC: Computing Power Evaluation Framework for Distributed Heterogeneous Intelligent Computing Systems

Computing Power Network (CPN) plays an important role as an infrastructure in various application fields. The evaluation of computing power in distributed heterogeneous intelligent computing systems serves as the basis of CPN and is especially crucial in the on-demand allocation of computing power resources and tasks scheduling. In this paper, we propose a computing power evaluation metric and computing Power Evaluation framework for distributed heterogeneous Intelligent Computing systems (PEIC). The computing power evaluation metric integrates software and hardware resource information of the devices into five primary metrics, as data processing capacity, data storage capacity, data transfer capacity, intelligent computing service capacity, and device computing resource utilization capacity. In the PEIC framework, we combine static resource information with dynamic intelligent computing service capacities. Entropy Weight Method (EWM) is used to solve the problem of collaborative influence between devices, and Analytic Hierarchy Process (AHP) method is used to give a comprehensive computation power evaluation for distributed heterogeneous intelligent computing systems. We also implemented the PEIC framework in a real environment and run five typical application scenarios such as image classification and natural language processing, and comprehensively evaluated the computation power of the system.