Operating cloud service infrastructures requires high energy efficiency while ensuring a satisfactory service level. Motivated by data centers, we consider a workload routing and server speed control policy applicable to the system operating under fluctuating demands. Dynamic control algorithms are generally more energy-efficient than static ones. However, they often require frequent information exchanges between routers and servers, making the data centers' management hesitate to deploy these algorithms. This study presents a static routing and server speed control policy that could achieve energy efficiency similar to a dynamic algorithm and eliminate the necessity of frequent communication among resources. We take a robust queueing theoretic approach to response time constraints for the quality of service (QoS) conditions. Each server is modeled as a G/G/1 processor sharing queue, and the concept of uncertainty sets defines the domain of stochastic primitives. We derive an approximative upper bound of sojourn times from uncertainty sets and develop an approximative sojourn time quantile estimation method for QoS. Numerical experiments confirm the proposed static policy offers competitive solutions compared with the dynamic algorithm.

翻译：运行云服务基础设施需要在确保满意服务水平的同时实现高能效。受数据中心启发，我们提出了一种适用于系统在波动需求下运行的工作负载路由与服务器速度控制策略。动态控制算法通常比静态算法更节能，然而，它们往往需要路由器和服务器之间频繁交换信息，这导致数据中心管理者在部署这些算法时犹豫不决。本研究提出了一种静态路由与服务器速度控制策略，它既能达到与动态算法相似的能效，又消除了资源间频繁通信的需求。我们采用鲁棒排队理论方法处理服务质量（QoS）条件下的响应时间约束。每个服务器被建模为G/G/1处理器共享队列，不确定性集合的概念定义了随机原语的定义域。我们从不确定性集合推导出逗留时间的近似上界，并开发了一种用于QoS的近似逗留时间分位数估计方法。数值实验证实，与动态算法相比，所提出的静态策略能提供具有竞争力的解决方案。