Scholarlink

With the surge in computationally intensive applications, supercomputing centers (SCs) must balance high-performance computing with sustainability amid dynamic energy markets and carbon policies. This paper presents the Cost-Effective Optimization Planning (CEOP) framework, which uses Mixed-Integer Linear Programming (MILP) to optimize long-term resource planning across distributed SCs. CEOP integrates time-variant electricity pricing, renewable energy availability, region-specific subsidies, and cross-regional data transmission into a unified model. Its multi-objective optimization minimizes operational costs, energy use, and carbon emissions while meeting service-level agreement (SLA) requirements, effectively balancing cost, sustainability, and performance. Simulations with real-world data show CEOP cuts annual operational costs by about 20% compared to traditional approaches, providing SC operators with actionable strategies for sustainable, large-scale computing infrastructure.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Computility Planning for Sustainable Supercomputing Centers

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jiayi Du,
  • Haodong Li,
  • Zheng Xiao,
  • Jiayi Liao

摘要

With the surge in computationally intensive applications, supercomputing centers (SCs) must balance high-performance computing with sustainability amid dynamic energy markets and carbon policies. This paper presents the Cost-Effective Optimization Planning (CEOP) framework, which uses Mixed-Integer Linear Programming (MILP) to optimize long-term resource planning across distributed SCs. CEOP integrates time-variant electricity pricing, renewable energy availability, region-specific subsidies, and cross-regional data transmission into a unified model. Its multi-objective optimization minimizes operational costs, energy use, and carbon emissions while meeting service-level agreement (SLA) requirements, effectively balancing cost, sustainability, and performance. Simulations with real-world data show CEOP cuts annual operational costs by about 20% compared to traditional approaches, providing SC operators with actionable strategies for sustainable, large-scale computing infrastructure.