Architectural Design and Key Technologies of Cross-Regional Heterogeneous Computing Resource Interconnection Platform

Bing Sun, Guohuan Song

Keywords: heterogeneous computing, computing power networks, computing power identifier, heterogeneous resource adapter, dynamic plugin management

TL;DR: This paper proposes a platform for seamless cross-regional interconnection of heterogeneous computing resources, using a granular identifier and a dynamic adapter for efficient resource management.

Abstract: The exponential growth in intelligent computing demand has created significant structural imbalances in computing power supply, characterized by geographical disparities, architectural heterogeneity, and management silos. This paper introduces a cross-regional heterogeneous computing resource interconnection platform based on a layered, decoupled cloud-native architecture. Its core contributions are two key technologies: a computing power identifier system that establishes a unified semantic framework for diverse resources, and a heterogeneous resource adapter that enables seamless integration through a dynamic, containerized plugin mechanism. The platform has been implemented and validated in a regional trial. Results demonstrate its capability for large-scale integration of heterogeneous resources, achieving a cross-domain scheduling success rate exceeding 99%. These findings validate the platform’s effectiveness in building a scalable, efficient, and reliable computing interconnection system, providing critical technical support for the evolution of future computing networks.

Submission Number: 54