Abstract not found

One of the key challenges for high-density servers (e.g., blades) is the increased costs in addressing the power and heat density associated with compaction. Prior approaches have mainly focused on reducing the heat generated at the level of an individual server. In contrast, this work proposes power efficiencies at a larger scale by leveraging statistical properties of concurrent resource usage across a collection of systems (“ensemble”). Specifically, we discuss an implementation of this approach at the blade enclosure level to monitor and manage the power across the individual blades in a chassis. Our approach requires low-cost hardware modifications and relatively simple software support. We evaluate our architecture through both prototyping and simulation. For workloads representing 132 servers from nine different enterprise deployments, we show significant power budget reductions at performances comparable to conventional systems. 1.

Power management has become increasingly necessary in large-scale datacenters to address costs and limitations in cooling or power delivery. This paper explores how to integrate power management mechanisms and policies with the virtualization technologies being actively deployed in these environments. The goals of the proposed VirtualPower approach to online power management are (i) to support the isolated and independent operation assumed by guest virtual machines (VMs) running on virtualized platforms and (ii) to make it possible to control and globally coordinate the effects of the diverse power management policies applied by these VMs to virtualized resources. To attain these goals, VirtualPower extends to guest VMs ‘soft ’ versions of the hardware power states for which their policies are designed. The resulting technical challenge is to appropriately map VM-level updates made to soft power states to actual changes in the states or in the allocation of underlying virtualized hardware. An implementation of VirtualPower Management (VPM) for the Xen hypervisor addresses this challenge by provision of multiple system-level abstractions including VPM states, channels, mechanisms, and rules. Experimental evaluations on modern multicore platforms highlight resulting improvements in online power management capabilities, including minimization of power consumption with little or no performance penalties and the ability to throttle power consumption while still meeting application requirements. Finally, coordination of online methods for server consolidation with VPM management techniques in heterogeneous server systems is shown to provide up to 34% improvements in power consumption.

Power densities have been increasing rapidly at all levels of server systems. To counter the high temperatures resulting from these densities, systems researchers have recently started work on software-based thermal management. Unfortunately, research in this new area has been hindered by the limitations imposed by simulators and real measurements. In this paper, we introduce Mercury, a software suite that avoids these limitations by accurately emulating temperatures based on simple layout, hardware, and componentutilization data. Most importantly, Mercury runs the entire software stack natively, enables repeatable experiments, and allows the study of thermal emergencies without harming hardware reliability. We validate Mercury using real measurements and a widely used commercial simulator. We use Mercury to develop Freon, a system that manages thermal emergencies in a server cluster without unnecessary performance degradation. Mercury will soon become available from

Abstract—High Performance Computing data centers have been rapidly growing, both in number and size. Thermal management of data centers can address dominant problems associated with cooling such as the recirculation of hot air from the equipment outlets to their inlets, and the appearance of hot spots. In this paper, we show through formalization that minimizing the peak inlet temperature allows for the lowest cooling power needs. Using a low-complexity, linear heat recirculation model, we define the problem of minimizing the peak inlet temperature within a data center through task assignment (MPIT-TA), consequently leading to minimal cooling requirement. We also provide two methods to solve the formulation, XInt-GA, which uses a genetic algorithm and XInt-SQP, which uses sequential quadratic programming. Results from small-scale data center simulations show that solving the formulation leads to an inlet temperature distribution that, compared to other approaches, is 2 °C to 5 °C lower and achieves about 20%-30 % cooling energy savings at common data center utilization rates. Moreover, our algorithms consistently outperform MinHR, a recirculation-reducing task placement algorithm in the literature.

Abstract — Recent advances have demonstrated the potential benefits of coordinated management of thermal load in data centers, including reduced cooling costs and improved resistance to cooling system failures. A key unresolved obstacle to the practical implementation of thermal load management is the ability to predict the effects of workload distribution and cooling configurations on temperatures within a data center enclosure. The interactions between workload, cooling, and temperature are dependent on complex factors that are unique to each data center, including physical room layout, hardware power consumption, and cooling capacity; this dictates an approach that formulates management policies for each data center based on these properties. We propose and evaluate a simple, flexible method to infer a detailed model of thermal behavior within a data center from a stream of instrumentation data. This data — taken during normal data center operation — includes continuous readings taken from external temperature sensors, server instrumentation, and computer room air conditioning units. Experimental results from a representative data center show that automatic thermal mapping can predict accurately the heat distribution resulting from a given workload distribution and cooling configuration, thereby removing the need for static or manual configuration of thermal load management systems. We also demonstrate how our approach adapts to preserve accuracy across changes to cluster attributes that affect thermal behavior — such as cooling settings, workload distribution, and power consumption. I.

Virtualization technology offers powerful resource management mechanisms, including performance-isolating resource schedulers, live migration, and suspend/resume. But how should networked virtual computing systems use these mechanisms? A grand challenge is to devise practical policies to drive these mechanisms in a self-managing or “autonomic” system, without relying on human operators. This paper explores architectural and algorithmic issues for resource management policy and orchestration in Shirako, a system for on-demand leasing of shared networked resources in federated clusters. Shirako enables a flexible factoring of resource management functions across the participants in a federated system, to accommodate a range of models of distributed virtual computing. We present extensions to Shirako to provision fine-grained virtual machine “slivers ” and drive virtual machine migration. We illustrate the interactions of provisioning and placement/migration policies, and their impact. 1

Abstract — The increasing costs of power delivery and cooling, as well as the trend toward higher-density computer systems, have created a growing demand for better power management in server environments. Despite the increasing interest in this issue, little work has been done in quantitatively understanding power consumption trends and developing simple yet accurate models to predict full-system power. We study the component-level power breakdown and variation, as well as temporal workload-specific power consumption of an instrumented power-optimized blade server. Using this analysis, we examine the validity of prior adhoc approaches to understanding power breakdown and quantify several interesting trends important for power modeling and management in the future. We also introduce Mantis, a nonintrusive method for modeling full-system power consumption and providing real-time power prediction. Mantis uses a onetime calibration phase to generate a model by correlating AC power measurements with user-level system utilization metrics. We experimentally validate the model on two server systems with drastically different power footprints and characteristics (a low-end blade and high-end compute-optimized server) using a variety of workloads. Mantis provides power estimates with high accuracy for both overall and temporal power consumption, making it a valuable tool for power-aware scheduling and analysis. I.

With power having become a critical issue in the operation of data centers today, there has been an increased push towards the vision of “energy-proportional computing”, in which no power is used by idle systems, very low power is used by lightly loaded systems, and proportionately higher power at higher loads. Unfortunately, given the state of the art of today’s hardware, designing individual servers that exhibit this property remains an open challenge. However, even in the absence of redesigned hardware, we demonstrate how optimization-based techniques can be used to build systems with off-the-shelf hardware that, when viewed at the aggregate level, approximate the behavior of energy-proportional systems. This paper explores the viability and tradeoffs of optimization-based approaches using two different case studies. First, we show how different power-saving mechanisms can be combined to deliver an aggregate system that is proportional in its use of server power. Second, we show early results on delivering a proportional cooling system for these servers. When compared to the power consumed at 100 % utilization, results from our testbed show that optimization-based systems can reduce the power consumed at 0 % utilization to 15 % for server power and 32 % for cooling power. 1

Thermal management has become a critical requirement for today’s power-dense server clusters, due to the negative impact of high temperatures on the reliability of computer hardware. Recognizing this fact, researchers have started to design software-based thermal management policies that leverage high-level information to control system-wide temperatures effectively. Unfortunately, designing these policies is currently a challenge, since it is difficult to predict the exact temperature and performance that would result from trying to react to a thermal emergency. Reactions that are excessively severe may cause unnecessary performance degradation and/or generate emergencies in other parts of the system, whereas reactions that are excessively mild may take relatively long to become effective (if at all), compromising the reliability of the system. To address this challenge, in this paper we propose C-Oracle, a software infrastructure for Internet services that dynamically predicts the temperature and performance impact of different thermal management reactions into the future, allowing the thermal management policy to select the best reaction at each point in time. C-Oracle makes predictions based on simple models of temperature, component utilization, and policy behavior that can be solved efficiently. We experimentally evaluate C-Oracle for thermal management policies based on load redistribution and dynamic voltage/frequency scaling in both single-tier and multi-tier services. Our results show that, regardless of management policy or service organization, C-Oracle enables non-trivial decisions that effectively manage thermal emergencies, while avoiding unnecessary performance degradation. 1