Data centers hosting internet-scale services consume megawatts of power. Mainly for cost reasons but also to appease environmental concerns, data center operators are interested to reduce their use of energy. This thesis investigates if and how hardware virtualization helps to improve the energy efficiency of modern cloud data centers. Our main motivation is to power off unused servers to save energy. The work encompasses three major parts: First, a simulation-driven analysis to quantify the benefits of known reservation times in infrastructure clouds. Virtual machines with similar expiration times are co-located to increase the probability to power down unused physical hosts. Second, we propose and prototyped a system to deliver truly on-demand cloud services. Idle virtual machines are suspended to free resources and as a first step to power off the physical server. Third, a novel block-level data synchronization tool enables fast and efficient state replication. Frequent state synchronization is necessary to prevent data unavailability: powering down a server disables access to the locally attached disks and any data stored on them. The techniques effectively reduce the overall number of required servers either through optimized scheduling or by suspending idle virtual machines. Fewer live servers translate into proportional energy savings, as the unused servers must no longer be powered.

Data centers hosting internet-scale services consume megawatts of power. Mainly for cost reasons but also to appease environmental concerns, data center operators are interested to reduce their use of energy. This thesis investigates if and how hardware virtualization helps to improve the energy efficiency of modern cloud data centers. Our main motivation is to power off unused servers to save energy. The work encompasses three major parts: First, a simulation-driven analysis to quantify the benefits of known reservation times in infrastructure clouds. Virtual machines with similar expiration times are co-located to increase the probability to power down unused physical hosts. Second, we propose and prototyped a system to deliver truly on-demand cloud services. Idle virtual machines are suspended to free resources and as a first step to power off the physical server. Third, a novel block-level data synchronization tool enables fast and efficient state replication. Frequent state synchronization is necessary to prevent data unavailability: powering down a server disables access to the locally attached disks and any data stored on them. The techniques effectively reduce the overall number of required servers either through optimized scheduling or by suspending idle virtual machines. Fewer live servers translate into proportional energy savings, as the unused servers must no longer be powered.

The concepts of 'data warehousing' and 'on-line analytical processing' have seen a growing interest in the research and commercial product community. Today, the trend moves away from complex centralized data warehouses to distributed data marts integrated in a common conceptual schema. However, as the first part of this paper demonstrates, there are many problems and little solutions for large distributed decision support systems in worldwide operating corporations. After showing the benefits and problems of the distributed approach, this paper outlines possibilities for achieving performance in distributed online analytical processing. Finally, the architectural framework of the prototypical distributed OLAP system CUBESTAR is outlined.

Data centres and similar server clusters consume a large amount of energy. However, not all consumed energy produces useful work. Servers consume a disproportional amount of energy when they are idle, underutilised, or overloaded. The effect of these conditions can be minimised by attempting to balance the demand for and the supply of resources through a careful prediction of future workloads and their efficient consolidation. In this paper we extend the cutting stock problem for consolidating workloads having stochastic characteristics. Hence, we employ the aggregate probability density function of co-located and simultaneously executing services to establish valid patterns. A valid pattern is one yielding an overall resource utilisation below a set threshold. We tested the scope and usefulness of our approach on a 16-core server with 29 different benchmarks. The workloads of these benchmarks have been generated based on the CPU utilisation traces of 100 real-world virtual machines which we obtained from a Google data centre hosting more than 32000 virtual machines. Altogether, we considered 600 different consolidation scenarios during our experiment. We compared the performance of our approach-system overload probability, job completion time, and energy consumption-with four existing/proposed scheduling strategies. In each category, our approach incurred a modest penalty with respect to the best performing approach in that category, but overall resulted in a remarkable performance clearly demonstrating its capacity to achieve the best trade-off between resource consumption and performance.