Technical Note—Cloud Cost Optimization: Model, Bounds, and Asymptotics
研究企业如何动态管理多种云资源以处理计算任务，提出一种渐近最优策略，平衡预留与按需资源的成本、性能及延迟，基于亚马逊云数据验证效果。
A Near-Optimal Capacity and Scheduling Policy for Cloud Users In “Cloud Cost Optimization: Model, Bounds, and Asymptotics,” Qu, Dawande, and Janakiraman study a long-term, dynamic resource-optimization problem for firms managing various cloud resources to process incoming computing tasks over time. Cloud resources vary in attributes, such as computing speed, memory, accelerators, and storage, tailored to different computing tasks, such as data warehousing, scientific computing, machine learning, and data processing. Firms can choose reserved resources for long-term commitments at lower costs or on-demand resources for flexibility at a higher price. Firms face three key trade-offs: the cost disparity between reserved and on-demand resources, the variation in resource attributes affecting performance and cost, and the challenge of balancing delay and resource costs. We propose an asymptotically optimal policy, demonstrating its effectiveness through a detailed numerical study based on Amazon Web Services data.