This paper presents an architecture for resource management in hosting centers, focusing on energy efficiency. The system, called Muse, uses an economic model where services "bid" for resources based on their performance needs. The system dynamically adjusts resource allocations to balance supply and demand, improving energy efficiency and adapting to varying loads. Experimental results show that Muse can reduce server energy usage by up to 29% for typical Web workloads. Muse is designed to manage server resources in a hosting center, where multiple services share a common hardware base. The system continuously monitors load and adjusts resource allocations to optimize performance and energy efficiency. It uses a reconfigurable switching infrastructure to direct request traffic to the most efficient servers. The system also responds to power supply disruptions or thermal events by degrading service in accordance with Service Level Agreements (SLAs). The system's core is an economic model that allows services to bid for resources based on their performance needs. This model enables adaptive resource provisioning in accordance with flexible SLAs, specifying dynamic tradeoffs of service quality and cost. Muse promotes energy efficiency by balancing the cost of resources (e.g., energy) against the benefit realized by employing them. The system automatically adjusts on-power capacity to scale with load, yielding significant energy savings for typical Internet service workloads. Muse's architecture includes a reconfigurable network switching fabric, load monitoring and estimation modules, and an executive that dynamically reallocates server resources and reconfigures the network to respond to variations in observed load, resource availability, and service value. The system's resource allocation algorithm, called Maximize Service Revenue and Profit (MSRP), uses an economic framework to manage resource allocation and provisioning in a way that maximizes resource efficiency and minimizes unproductive costs. The system's ability to dynamically adjust server allocations enables it to improve energy efficiency by matching a cluster's energy consumption to the aggregate request load and resource demand. Energy-conscious provisioning configures switches to concentrate request load on a minimal active set of servers for the current aggregate load level. Active servers always run near a configured utilization threshold, while the excess servers transition to low-power idle states to reduce the energy cost of maintaining surplus capacity during periods of light load. The system's economic model allows for flexible resource allocation based on service demand and performance needs. The system's resource allocation algorithm, MSRP, uses an economic framework to manage resource allocation and provisioning in a way that maximizes resource efficiency and minimizes unproductive costs. The system's ability to dynamically adjust server allocations enables it to improve energy efficiency by matching a cluster's energy consumption to the aggregate request load and resource demand.This paper presents an architecture for resource management in hosting centers, focusing on energy efficiency. The system, called Muse, uses an economic model where services "bid" for resources based on their performance needs. The system dynamically adjusts resource allocations to balance supply and demand, improving energy efficiency and adapting to varying loads. Experimental results show that Muse can reduce server energy usage by up to 29% for typical Web workloads. Muse is designed to manage server resources in a hosting center, where multiple services share a common hardware base. The system continuously monitors load and adjusts resource allocations to optimize performance and energy efficiency. It uses a reconfigurable switching infrastructure to direct request traffic to the most efficient servers. The system also responds to power supply disruptions or thermal events by degrading service in accordance with Service Level Agreements (SLAs). The system's core is an economic model that allows services to bid for resources based on their performance needs. This model enables adaptive resource provisioning in accordance with flexible SLAs, specifying dynamic tradeoffs of service quality and cost. Muse promotes energy efficiency by balancing the cost of resources (e.g., energy) against the benefit realized by employing them. The system automatically adjusts on-power capacity to scale with load, yielding significant energy savings for typical Internet service workloads. Muse's architecture includes a reconfigurable network switching fabric, load monitoring and estimation modules, and an executive that dynamically reallocates server resources and reconfigures the network to respond to variations in observed load, resource availability, and service value. The system's resource allocation algorithm, called Maximize Service Revenue and Profit (MSRP), uses an economic framework to manage resource allocation and provisioning in a way that maximizes resource efficiency and minimizes unproductive costs. The system's ability to dynamically adjust server allocations enables it to improve energy efficiency by matching a cluster's energy consumption to the aggregate request load and resource demand. Energy-conscious provisioning configures switches to concentrate request load on a minimal active set of servers for the current aggregate load level. Active servers always run near a configured utilization threshold, while the excess servers transition to low-power idle states to reduce the energy cost of maintaining surplus capacity during periods of light load. The system's economic model allows for flexible resource allocation based on service demand and performance needs. The system's resource allocation algorithm, MSRP, uses an economic framework to manage resource allocation and provisioning in a way that maximizes resource efficiency and minimizes unproductive costs. The system's ability to dynamically adjust server allocations enables it to improve energy efficiency by matching a cluster's energy consumption to the aggregate request load and resource demand.