Data Center Power Research Articles

Negotiation game for joint IT and energy management in green datacenters

As the power demand of datacenters is increasing sharply, a promising solution is to power datacenters locally by renewable energies. However, one of the main challenges when operating such green datacenters is to conciliate the intermittent power supply and the power demand. To deal with this problem, we view the green datacenter as two sub-systems, namely, Information Technology (IT) sub-system which consumes energy, and electrical sub-system which supplies energy. The objective is to find an efficient trade-off between the power demand and power supply, respecting the operational requirements of both sub-systems (i.e., the requirements on utility, or monetary gain, which includes monetary revenue and monetary cost). First, we analyze the problem by a black-box approach. In this approach, the models of the two sub-systems are unknown to each other, and the two sub-systems negotiate by exchanging their power preferences. However, we found that the black-box approach cannot guarantee stable solutions in term of execution time and generational distance (which is the distance between a solution and the Pareto front). Then we introduce a semi black-box approach, in which the two sub-systems are modeled as the buyer and the supplier in a buyer–supplier negotiation game. We propose an algorithm that allows the buyer and supplier to negotiate, seeking for an efficient trade-off between the power demand and power supply. The analytical results show that the semi black-box algorithm converges to equilibrium, and these results are then confirmed by experimental results. We conduct the experiments by implementing a middleware of a datacenter powered by renewable energies. The experimental results show that the semi black-box algorithm improves significantly the stability, quality of service (QoS) and utility of the datacenter, compared to other algorithms. In term of stability, compared to the black-box algorithm, the semi black-box algorithm reduces the standard deviation of execution time and generational distance by 23 and 27 times, respectively. In term of QoS and utility, the semi black-box algorithm outperforms the algorithms that do not consider joint IT-energy management, as well as the algorithms that do not utilize a semi black-box design.

A parametric CFD study of computer room air handling bypass in air-cooled data centers

Computer room air handling (CRAH) bypass (BP) method is an application that reduces the total fan power in air-cooled data centers by lowering CRAH fan speed. BP fans supply the complementary fraction of server airflow from warmer room air through lower-resistance floor openings into the plenum. Hence, chillers need to operate at lower temperatures less efficiently, which poses an optimization problem. The reduced-order modeling tools in the literature adequately predict fan power. However, they assume well-mixed temperature at rack inlets to deal with this optimization problem and fail to quantify the impact of ignoring temperature non-uniformities. This study includes an experimental verification of computational fluid dynamics (CFD) modeling for CRAH BP method in predicting temperatures in a data center test-cell. The subsequent parametric study uses both CFD and reduced-order modeling tools on a more representative quadrant of a large data center to investigate induced and forced CRAH BP in both enclosed and open aisle configurations as well as comparing various plenum heights, IT load and server airflow rate. Results indicate where reduced-order modeling tools perform reasonably well and where the need for CFD emerges and identify various favorable data center designs and operating conditions for CRAH BP method.

The effect of server energy proportionality on data center power oversubscription

Modern data centers improve resource utilization with power oversubscription. The power hierarchy in a data center is oversubscribed by installing more servers than allowed by the power budget based on server peak power consumption. Power oversubscription is possible due to the statistically low likelihood of simultaneous peak power operation of multiple servers. As future servers become more energy proportional, the opportunity for greater power oversubscription increases. The challenge is to quantify the level of oversubscription that can be attained. In this paper, we quantify the level of oversubscription possible for a given acceptable probability of power overload for servers characterized by energy proportionality metric and workload distribution. We develop a theoretical framework to characterize and predict the relationship between server energy proportionality and power oversubscription. We verify our framework through an extensive empirical study using publicly available SPECpower benchmark data for over 500 server models and publicly available Google cluster utilization data. Using our framework, a data center operator can predict the additional power oversubscription possible when replacing existing servers with a newer model of more energy proportional servers.

Efficient Algorithms for Approximate Single-Source Personalized PageRank Queries

Given a graph G , a source node s, and a target node t , the personalized PageRank ( PPR ) of t with respect to s is the probability that a random walk starting from s terminates at t . An important variant of the PPR query is single-source PPR ( SSPPR ), which enumerates all nodes in G and returns the top- k nodes with the highest PPR values with respect to a given source s . PPR in general and SSPPR in particular have important applications in web search and social networks, e.g., in Twitter’s Who-To-Follow recommendation service. However, PPR computation is known to be expensive on large graphs and resistant to indexing. Consequently, previous solutions either use heuristics, which do not guarantee result quality, or rely on the strong computing power of modern data centers, which is costly. Motivated by this, we propose effective index-free and index-based algorithms for approximate PPR processing, with rigorous guarantees on result quality. We first present FORA, an approximate SSPPR solution that combines two existing methods—Forward Push (which is fast but does not guarantee quality) and Monte Carlo Random Walk (accurate but slow)—in a simple and yet non-trivial way, leading to both high accuracy and efficiency. Further, FORA includes a simple and effective indexing scheme, as well as a module for top- k selection with high pruning power. Extensive experiments demonstrate that the proposed solutions are orders of magnitude more efficient than their respective competitors. Notably, on a billion-edge Twitter dataset, FORA answers a top-500 approximate SSPPR query within 1s, using a single commodity server.

GASP: Genetic Algorithms for Service Placement in Fog Computing Systems

Fog computing is becoming popular as a solution to support applications based on geographically distributed sensors that produce huge volumes of data to be processed and filtered with response time constraints. In this scenario, typical of a smart city environment, the traditional cloud paradigm with few powerful data centers located far away from the sources of data becomes inadequate. The fog computing paradigm, which provides a distributed infrastructure of nodes placed close to the data sources, represents a better solution to perform filtering, aggregation, and preprocessing of incoming data streams reducing the experienced latency and increasing the overall scalability. However, many issues still exist regarding the efficient management of a fog computing architecture, such as the distribution of data streams coming from sensors over the fog nodes to minimize the experienced latency. The contribution of this paper is two-fold. First, we present an optimization model for the problem of mapping data streams over fog nodes, considering not only the current load of the fog nodes, but also the communication latency between sensors and fog nodes. Second, to address the complexity of the problem, we present a scalable heuristic based on genetic algorithms. We carried out a set of experiments based on a realistic smart city scenario: the results show how the performance of the proposed heuristic is comparable with the one achieved through the solution of the optimization problem. Then, we carried out a comparison among different genetic evolution strategies and operators that identify the uniform crossover as the best option. Finally, we perform a wide sensitivity analysis to show the stability of the heuristic performance with respect to its main parameters.

Stochastic Models for Optimizing Availability, Cost and Sustainability of Data Center Power Architectures through Genetic Algorithm

In recent years, the growth of information technology has required higher reliability, accessibility, collaboration, availability, and a reduction of costs on data centers due to factors such as social network, cloud computing, and e-commerce. These systems require redundant mechanisms on the data center infrastrucutre to achieve high availability, which may increase the electric energy consumption, impacting in both the sustainability and cost. This work proposes a multi-objective optimization approach, based on Genetic Algorithms, to optimize cost, sustainability and availability of data center power infrastructures. The main goal is to maximize availability and minimize cost and exergy consumed (adopted to estimate the environmental impacts). In order to compute such metrics, this work adopts the energy flow model (EFM), reliability block diagrams (RBD) and stochastic petri nets (SPN). Two case studies are conducted to show the applicability of the proposed strategy: (i) takes into account 5 typical data center architectures that were optimized to conduct the validation process of the proposed strategy; (ii) uses the optimization strategy in two architectures classified by ANSI / TIA-942 (TIER I and II). In both case studies, significant improvements were achieved in the results, which were very close to the optimum one that was obtained by a brute force algorithm that analyzes all the possibilities and returns the optimal solution. It is worth mentioning that the time used to obtain the results using the genetic algorithm approach was significantly lower (6,763,260 times), in comparison with the strategy which combines all the possible combinations to obtain the optimal result.

A novel optimized approach for resource reservation in cloud computing using producer–consumer theory of microeconomics

Designing economic pricing mechanisms have recently attracted a great deal of attention in the context of cloud computing. We believe that microeconomics theory is a good candidate to model the resource reservation operations in cloud networks. Producer–consumer theory of microeconomics guarantees the maximization of social welfare of the customers, conditional that the particular consideration concerning customers and producers are met. As is the case in real-world cloud datacenters, the workload associated with each user is fed into the system and then the user is bound to a virtual machine (VM). In this research, we propose a microeconomic-inspired resource reservation scheme for cloud computing. The designed mechanism includes two steps: in the first step, we seek to find a Pareto efficient reservation set concerning bandwidth of VMs, and in the second step, our goal is to place VMs’ reserved bandwidth rates on physical hosts. In our modeling, VMs and the cloud network are considered as consumers and producers of the market, respectively. Also, the bandwidth of requested services is considered as commodity. As is the case in microeconomics, we prove that the aggregation of users’ utilities (users’ social welfare in microeconomics terminology) could reach to global maximum, known as Pareto efficiency. After finding the best set of reserved bandwidth rates in the first step of mechanism, in the second step, the mechanism seeks to find the best placement for VMs on physical hosts. The placement operation is performed in such a way that results in minimization of total consumed power in datacenter. Since the VM placement problem has been proven to be NP-hard, we use a metaheuristic cuckoo search optimization approach to solve the optimization problem. Simulation results, obtained through the CloudSim framework, established the robustness of the proposed method in terms of significant criteria such as users’ welfare, consumed power and Pareto optimality.

Influence of cooling architecture on data center power consumption

Almost thirty percent of the power consumed by data centers (DCs) is attributable to the cooling of IT equipment (ITE). There are opportunities to reduce a DC's energy budget by considering alternatives to traditional cooling methods, which experience inherent airflow deficiencies due to hot air recirculation and cold air bypass. Minimizing these two air distribution problems results in more effective cooling, but the two effects are manifest differently in the three conventional DC cooling architectures, i.e., (a) room-based, (b) row-based, and (c) rack-based cooling. Despite the intuitive logic that predicts improved cooling air distribution within row- and rack-based architectures that include shorter airflow pathlengths compared to room-based systems that have longer paths, the mechanism through which improvements translate into energy savings is not well understood. Therefore, we present methodologies that resolve the characteristic airflow and temperature distributions for three cooling architectures using computational fluid dynamics. These results inform thermodynamics models of the power consumptions that are required to cool these three architectures. The analysis reveals that row- and rack-based architectures reduce cooling power by much as 29% over a room-based architecture. Adding an enclosure within row- and rack-based architectures to separate the hot and cold airflows provides further 18% reduction in cooling power. This analysis facilitates better DC design from a cooling power consumption perspective.

Fog Computing for Smart Grid Systems in the 5G Environment: Challenges and Solutions

Currently, the demand for electricity is increasing day by day, which necessitates upgrading of the existing power grid system. The conventional power grid has already been replaced with modern ICT-based infrastructure, which is known as smart grid (SG). In SG, smart meters generate a huge amount of data, and it is a challenging task to store, process, and analyze the data, which varies with respect to volume, velocity, and variety. The data generated in an SG system is generally stored and analyzed using cloud computing (CC), which provides real-time response for various applications. However, to handle the latency issue during the data analytics in SG, fog computing (FC) has emerged as a new technology that provides most of the computing resources in proximity of the end users. It acts as a bridge between SG and CC to fill the gap between processing power of remote data centers and smart devices in SG systems. To handle the aforementioned issues, there is a requirement to set up advanced sensors and measurement systems having communication network backbones in the upcoming fifth generation (5G). In this article, we discuss the architecture of SG in the context of FC for making the decision about energy requirements by the smart devices at the fog layer. Moreover, the communication and computing aspects are also explored in the context of 5G network infrastructure. We examine the influence of FC on response time, transmission delay, and energy management costs for delay-sensitive applications.

Frequency regulation service provision in data center with computational flexibility

The rapid adoption of cloud storage and computing services led to unprecedented growth of data centers in the world. As bulk energy consumers, large-scale data centers in the U.S. rack up billions in electricity costs annually. Fortunately, the operational flexibility of data centers can be leveraged to provide valuable frequency regulation services in smart grids to mitigate the indeterminacy of the renewable generation resources. Specifically, this paper aims to leverage computational flexibility provided by servers, such as dynamic voltage frequency scaling and dummy loads. This paper develops a comprehensive framework for data center’s frequency regulation service provision in both hour-ahead market and real-time operations. A risk constrained hour-ahead bidding strategy along with a real-time data center power consumption control algorithm are developed to minimize electricity bills and the total response time of the requests. The introduction of dummy load, realistic bi-linear server power consumption model, and probabilistic forecast of electricity and frequency regulation service prices enable the data center to accurately follow frequency regulation signals, while reducing the financial risks associated with electricity market participation. The simulation results show that the proposed frequency regulation provision framework results not only in significant cost reduction for data centers, but also limits degradation in quality of service. Meanwhile, the stability and reliability of a power grid will be improved by the frequency regulation service provision.

A decision support system for waste heat recovery and energy efficiency improvement in data centres

Data centre sector is emerging as one of the largest and fastest growing industrial sectors, accounting for 3% of the global electricity supply and contributing to 4% of total greenhouse gas emissions. A framework for waste heat energy recovery and its accompanying decision support system is presented with the ability to evaluate the waste heat (source) and potential demand (sink) compatibility, perform exergy and temporal availability analysis, and undertake cost-benefits and environmental impact analysis of available heat recovery technologies. This four stages framework is implemented through the case study of a medium-sized commercial data centre to provide evidence of its applicability. The results demonstrated that the framework and decision support system can deliver a streamlined and optimised heat recovery strategy for reducing overall energy requirement in the data centre. About 68% of the inevitable waste heat from IT equipment can be recovered by the recommended solution, allowing the recovered waste heat in the form of warm water to be fed back into the facility for applications such as space heating, achieving about 10% improvement in data centre power usage effectiveness. The solutions given by the decision support system are generally accessible for most data centres as the technologies are readily available and often lead to short payback time and substantial energy savings.

Analyzing the performance/power tradeoff of the rCUDA middleware for future exascale systems

The computing power of supercomputers and data centers has noticeably grown during the last decades at the cost of an ever increasing energy demand. The need for energy (and power) of these facilities has finally limited the evolution of high performance computing, making that many researchers are concerned not only about performance but also about energy efficiency. However, despite the many concerns about energy consumption, the search for computing power continues. In this regard, the research on exascale systems, able to deliver 1018 floating point operations per second, has reached a widely consensus that these systems should operate within a maximum power budget of 20 megawatts. Many efficiency improvements are necessary for achieving this goal. One of these improvements is the usage of ARM low-power processors, as the Mont-Blanc project proposes. In this paper we analyze the combined use of ARM processors with the rCUDA remote GPU virtualization middleware as a way to improve efficiency even more. Results show that it is possible to speed up applications by almost 8x while reducing energy consumption up to 35% when rCUDA is used to access high-end GPUs. These improvements are achieved while maintaining a feasible average power consumption level for future exascale systems.

Study on identifying the vulnerability of the data center power network

With the wide application of the internet and rapid development of the information technology, the scale of the internet data center power network is growing larger and larger. Some of the troubles in the internet data center (IDC) power network may cause extensive service out of service or even collapse. However, without considering the characteristics of the IDC power network, the important node identification methods for the IDC power network based on graph mainly focus on the graph structure itself. Since a large number of automatic switching equipment such as ATS, STS and loop switch have been used in IDC, when the electrical fault occurs, the load will be automatically migrated to the adjacent electrical equipment, which may lead to overloading or even tripping. So the failure is extended. So how can we efficiently and effectively capture the fault evolution mechanism of the IDC power network? How do we go about identifying the vulnerable nodes causing the collapse of the IDC power network.In this article, we propose DCPNFEM, a novel fault evolution model for the IDC power networks, as well as a fitting algorithm, BCBL, which can solve the proposed problems. Our algorithms have the following properties: (a) Intuitively: it detects fault evolution mechanism, such as power load migration, electrical equipment tripping, and so on; (b) Timely: our method is based on the real-time power loads; (c) General: our algorithms are general and practical, which can be used in various power network topologies, including data center infrastructures from tier1 to tier4.Extensive experiments on a real IDC power network demonstrate that some particular nodes’ failure can lead to the power network crash under DCPNFEM. And BCBL algorithm outperforms better accuracy and speed than many other algorithms.

<italic>LCLC</italic> Converter With Optimal Capacitor Utilization for Hold-Up Mode Operation

In data center and telecommunication power supplies, the front-end dc–dc stage is required to operate with a wide input voltage range to provide hold-up time when ac input fails. Conventional LLC converter serving as the dc–dc stage is not suitable for this requirement, as the normal operation efficiency (at 400 V input) will be penalized once the converter is designed to achieve high peak gain (wide input voltage range). This paper examined the operation of the LCLC converter and revealed that the LCLC converter could be essentially equivalent to a set of LLC converters with different magnetizing inductors that are automatically adjusted for different input voltages. In nominal 400 V input operation, the LCLC converter behaves like an LLC converter with large magnetizing inductor, thus the resonant current is small. In the hold-up period, when the input voltage reduces, the equivalent magnetizing inductor will reduce together with switching frequency reducing, thus the converter achieves high peak gain. In this paper, a new design methodology is also proposed to achieve optimal utilization of the two resonant capacitors for high power application. To verify the effectiveness of the LCLC converter for hold-up operation, comprehensive analysis has been conducted; a detailed step by step design example based on capacitor voltage stress is introduced; an experimental LCLC prototype optimized at 400 V, with input voltage range of 250–400 V and 12 V/500 W as output has been presented.

A Survey on Power Management Techniques for Oversubscription of Multi-Tenant Data Centers

Power management for data centers has been extensively studied in the past 10 years. Most research has focused on owner-operated data centers with less focus on Multi-Tenant Data Centers (MTDC) or colocation data centers. In an MTDC, an operator owns the building and leases out space, power, and cooling to tenants to install their own IT equipment. MTDC’s present new challenges for data center power management due to an inherent lack of coordination between the operator and tenants. In this article, we conduct a comprehensive survey of existing MTDC power management techniques for demand response programs, sustainability, and/or power hierarchy oversubscription. Power oversubscription is of particular interest, as it can maximize resource utilization, increase operator profit, and reduce tenant costs. We create a taxonomy to classify and compare key works. Our taxonomy and review differ from existing works in that our emphasis is on safe power oversubscription, which has been neglected in previous surveys. We propose future research for prediction and control of power overload events in an oversubscribed MTDC.

High‐power‐density high‐efficiency LLC converter with an adjustable‐leakage‐inductance planar transformer for data centers

LLC converters can achieve high efficiency, high- power density, wide gain range, and galvanic isolation. This study presents a high-power-density and high-efficiency LLC converter with an adjustable-leakage-inductance planar transformer for the DC-DC conversion in data center power supplies. A planar transformer with adjustable leakage inductance is modelled and developed to provide an accurate design for resonant frequency and voltage gain. A detailed power loss analysis is performed. Furthermore, a thermal model is provided to aid in heat sink design. GaN devices are used in the converter to achieve high-switching frequency and low-switching loss. A 450 kHz, 360-400 V DC input, 12 V DC output, 750 W LLC converter prototype with an adjustable-leakage-inductance planar transformer is built and tested. The power density of the system is 15.4 W/cm 3 including an auxiliary power supply. The peak efficiency of the overall system is over 95%.

Heterogeneity Aware Workload Management in Distributed Sustainable Datacenters

The tremendous growth of cloud computing and large-scale data analytics highlight the importance of reducing datacenter power consumption and environmental impact of brown energy. While many Internet service operators have at least partially powered their datacenters by green energy, it is challenging to effectively utilize green energy due to the intermittency of renewable sources, such as solar or wind. We find that the geographical diversity of internet-scale services can be carefully scheduled to improve the efficiency of applying green energy in datacenters. In this paper, we propose a holistic heterogeneity-aware cloud workload management approach, sCloud, that aims to maximize the system goodput in distributed self-sustainable datacenters. sCloud adaptively places the transactional workload to distributed datacenters, allocates the available resource to heterogeneous workloads in each datacenter, and migrates batch jobs across datacenters, while taking into account the green power availability and QoS requirements. We formulate the transactional workload placement as a constrained optimization problem that can be solved by nonlinear programming. Then, we propose a batch job migration algorithm to further improve the system goodput when the green power supply varies widely at different locations. Finally, we extend sCloud by integrating a flexible batch job manager to dynamically control the job execution progress without violating the deadlines. We have implemented sCloud in a university cloud testbed with real-world weather conditions and workload traces. Experimental results demonstrate sCloud can achieve near-to-optimal system performance while being resilient to dynamic power availability. sCloud with the flexible batch job management approach outperforms a heterogeneity-oblivious approach by 37 percent in improving system goodput and 33 percent in reducing QoS violations.

EnergyQARE

Power market operators have recently introduced smart grid demand response (DR), in which electricity consumers regulate their power usage following market requirements. DR helps stabilize the grid and enables integrating a larger amount of intermittent renewable power generation. Data centers provide unique opportunities for DR participation due to their flexibility in both workload servicing and power consumption. While prior studies have focused on data center participation in legacy DR programs such as dynamic energy pricing and peak shaving, this article studies data centers in emerging DR programs, i.e., demand side capacity reserves. Among different types of capacity reserves, regulation service reserves (RSRs) are especially attractive due to their relatively higher value. This article proposes EnergyQARE , the Energy and Q uality-of-Service (QoS) A ware R SR E nabler, an approach that enables data center RSR provision in real-life scenarios. EnergyQARE not only provides a bidding strategy in RSR provision, but also contains a runtime policy that adaptively modulates data center power through server power management and server provisioning based on workload QoS feedback. To reflect real-life scenarios, this runtime policy handles a heterogeneous set of jobs and considers transition time delay of servers. Simulated numerical results demonstrate that in a general data center scenario, EnergyQARE provides close to 50% of data center average power consumption as reserves to the market and saves up to 44% in data center electricity cost, while still meeting workload QoS constraints. Case studies in this article show that the percentages of savings are not sensitive to a specific type of non-interactive workload, or the size of the data center, although they depend strongly on data center utilization and parameters of server power states.

A First Look at Thermal Attacks in Multi-Tenant Data Centers

This paper focuses on the emerging threat of thermal attacks in a multi-tenant data center. It discovers that a malicious tenant (i.e., attacker) can inject additional thermal loads beyond the shared cooling system capacity, thus resulting in overheating and possible system downtime. Importantly, the attacker can launch thermal attacks in a stealthy manner by discharging batteries inside its servers and still keeping its power drawn from the data center power distribution system under its subscribed capacity.

Will surface effects dominate in quasi-two-dimensional gallium oxide for electronic and photonic devices?

The ultra-wide band gap semiconductor Ga2O3 has advantages for power electronics applications in the automotive industry, data center power management and industrial systems but attention must be paid to its surface sensitivity to the environment.