Environmental sustainability is an increasingly important design constraint for next-generation servers and datacenters. Unlike prior studies that focus on operational energy use, we study the environmental impact of current designs across the entire lifecycle, including embedded impact factors related to material use and manufacturing. Based on the insights provided by this study, we propose a solution co-designed across system architecture and physical packaging, including (1) material-efficient physical organization, (2) environmentally-efficient cooling infrastructures, and (3) effective design of system architectures to reuse components — all working together to improve sustainability. We provide a detailed evaluation of our proposed solution in terms of sustainability, thermal manageability, and computational performance. Our results show that the proposed approach is effective in addressing the (often non-intuitive) tradeoffs between performance and different components of sustainability.

In data centers with raised floor architecture, the floor tiles are typically perforated, delivering the cold air from the plenum to the inlets of equipment located in racks. The environment of these data centers is dynamic in that the workload and power dissipation fluctuate considerably over both short-term and long-term time scales. As such, airflow requirements vary continuously. However, due to labor costs and lack of expertise, the tiles are adjusted infrequently, and many data centers are grossly over provisioned for airflow in general and/or lack sufficient airflow delivery in certain local areas. This wastes energy and reduces data center thermal capacity. We have previously introduced Kratos, an Adaptive Vent Tile (AVT) technology that addresses this problem by automatically adjusting mechanical louvers mounted to the tiles in response to the needs of nearby IT equipment. Our initial results were limited to a 3-tile test bed that allowed us to prove concept but did not provide for scalability. This paper extends the previous work by expanding the size of the test bed to 28 tiles and 29 racks located in multiple thermal zones. We present experimental modeling results on the MIMO (Multi-Input Multi-Output) system and provide insights on the external behavior of the system through CFD (Computational Fluid Dynamic) analysis. We develop an MPC (Model-based Predictive Control) controller to maintain the temperatures of racks below the thresholds through vent tile tuning. Experimental results show that the controller can maintain the temperature below the thresholds while reducing overall cooling air requirements.

A significant amount of energy consumption is now attributed to data centers due to their ever increasing numbers, size and power densities. Thus, there are efforts focused at making a data center more sustainable by reducing its energy consumption and carbon footprint. This requires an end-to-end management approach with requirements derived from service level agreements (SLAs) and a flexible infrastructure that can be closely monitored and finely controlled. The infrastructure can then be manipulated to satisfy the requirements while optimizing for sustainability metrics and total cost of operations. In this paper, we explore the role of data analysis, visualization and knowledge discovery techniques in improving the sustainability of a data center. We present use cases from a large, sensor-rich, state-of-the-art data center on the application of these techniques to the three main sub-systems of a data center, namely, power, cooling and compute. Furthermore, we provide recommendations for where these techniques can be used within these sub-systems for improving sustainability metrics of a data center.

Abstract-Efficient and reliable operation of today's data centers, which host IT equipment with ever-increasing power density, relies heavily on the cooling system to meet the thermal management needs of the IT equipment with minimal environmental footprint. The dynamic IT workload, together with the spatial variance of cooling efficiencies, creates both temporal and spatial non-uniformities within the data centers. Most data centers use zonal cooling actuators, such as computer room air conditioners (CRAC), to alleviate the local "hot spots". Without proper localized cooling actuation mechanisms, the cooling capacity is usually over-provisioned that leads to waste of energy. To address this problem, we introduce adaptive vent tiles (AVT) for local cooling adjustment, and develop a holistic multivariable model based on the mass and energy balance principles to capture the effects of both zonal and local cooling actuation on the inlet temperatures of the racks that host the IT equipment. A model predictive controller is then proposed to minimize the total cooling power while meeting the thermal requirements of the racks. The zonal and local cooling actuation is coordinated in such a unified framework for the provisioning, transport and distribution of the cooling resources in the data centers. The proposed holistic cooling approach is validated in a production data center. Experimental results indicate that up to 36% of CRAC units blower power can be saved, compared with the state of the art control solution.

Data centers are the computational hub of the next generation. Rise in demand for computing has driven the emergence of high density datacenters. With the advent of high density, mission-critical datacenters, demand for electrical power for compute and cooling has grown. Deployment of a large number of high powered computer systems in very dense configurations in racks within data centers will result in very high power densities at room level. Hosting business and mission-critical applications also demand a high degree of reliability and flexibility. Managing such high power levels in the data center with cost effective reliable cooling solutions is essential to feasibility of pervasive compute infrastructure. Energy consumption of data centers can also be severely increased by over-designed air handling systems and rack layouts that allow the hot and cold air streams to mix. Absence of rack level temperature monitoring has contributed to lack of knowledge of air flow patterns and thermal management issues in conventional data centers. In this paper, we present results from exploratory data analysis (EDA) of rack-level temperature data collected over a period of several months from a conventional production datacenter. Typical datacenters experience surges in power consumption due to rise and fall in compute demand. These surges can be long term, short term or periodic, leading to associated thermal management challenges. Some variations may also be machine-dependent and vary across the datacenter. Yet other thermal perturbations may be localized and momentary. Random variations due to sensor response and calibration, if not identified, may lead to erroneous conclusions and expensive faults. Among other indicators, EDA techniques also reveal relationships among sensors and deployed hardware in space and time. Identification of such patterns can provide significant insight into data center dynamics for future forecasting purposes. Knowledge of such metrics enables energy-efficient thermal management by helping to create strategies for normal operation and disaster recovery for use with techniques like dynamic smart cooling.