Practices of a Computing Service Platform For Advanced Computing and AI Applications

Qing Guo, Chief Engineer at Sugon, presents the company's experience building and operating a large-scale computing service platform for advanced computing and AI applications. The platform connects heterogeneous supercomputing and intelligent-computing data centers across regions into a unified fabric through a three-layer architecture: a computing resource layer interconnecting data centers, a scheduling and management layer providing distributed heterogeneous resource scheduling and container deployment, and an application service layer offering standard APIs, SaaS applications, AI models, and datasets. Supporting elements include an application marketplace, professional services for porting and optimizing software on heterogeneous hardware, and an ecosystem connecting computing centers, service providers, developers, and users. Guo cites key technology work on million-scale user concurrency, hyper-scale job scheduling, cross-region multi-cluster scheduling, and distributed billing, and reports operational figures of more than one million users, roughly 215,000 tasks per day, and over 196 million tasks handled since launch. A highlighted development is the SuperAC intelligent agent for scientific computing, which takes natural-language task descriptions and autonomously plans workflows, selects resources, submits jobs, and generates reports, reducing tasks that took a day to about an hour and supporting over 100 scientific computing scenarios. In the Q&A, Guo describes a multi-agent architecture using MCP to capture domain knowledge, access via web, desktop, and mobile clients, and jurisdiction-dependent handling of workloads, with customers mainly in China and expanding to Hong Kong and Singapore.