Data Center Power Research Articles

Data Center Power and Energy Management: Past, Present, and Future

Data Center Power and Energy Management: Past, Present, and Future

AI-powered data center power control for sustainable development

AI-powered data center power control for sustainable development

OCL-MEC: An online CPU-core prediction based on load balancing framework for offloading resource management in mobile edge computing environment

Clients can increase or decrease the number of resources they use dynamically over time due to the elasticity of cloud resources. As a result, variations in resource demands and predefined VM sizes result in a lack of resource utiliation, load imbalances, and excessive power consumption. A framework of efficient resource management is proposed to address these issues, balancing the load accordingly and anticipating the resource utilization of the servers. By optimizing resource utilization and minimizing the number of active servers, this technique facilitates power savings. Under/overloaded servers reduce energy consumption, execution delay, and performance degradation through a resource prediction system that is deployed at the CPU. Moreover, OCL-MEC load-balancing and resource allocation algorithms are proposed to reduce data center network traffic and power consumption. Experiments on real-world workload datasets, namely Bitsbrain VM traces, are conducted to evaluate the proposed framework. Different performance metrics demonstrate the superiority of the proposed framework over state-of-the-art approaches. Power savings of up to 98 % can be achieved by the OCL-MEC framework using a decision tree load balancing model based on HMM prediction systems.

Exploring the Evolution Trend of China’s Digital Carbon Footprint: A Simulation Based on System Dynamics Approach

The rapid growth of the digital economy has heightened concerns over its environmental impacts, particularly in terms of carbon dioxide emissions. In contrast to previous studies that focus on the positive effects of digital technology on reducing carbon emissions, this paper provides a detailed analysis of the various factors that influence digital economy carbon emissions and their interrelationships, using the system dynamics method to simulate and predict China’s future digital economy carbon emission baseline from 2016 to 2046. Four different scenarios were established by adjusting parameters for the percentage of the digital economy, e-waste growth rate, and data center power consumption. The simulation results indicate the following: (1) The baseline scenario shows China’s digital economy carbon emissions peaking at 1.9045 billion tons in 2041 after an initial increase and subsequent decrease. (2) Single-policy simulations indicate that changing the digital economy’s scale and e-waste growth independently leads to peak emissions of 1.9205 billion tons and 1.5525 billion tons, respectively. Adjusting data center power consumption has a greater impact, increasing the peak emissions to 2.1675 billion tons, a 13.82% rise from the baseline. (3) Under the comprehensive regulation scenario, emissions peak in 2040 at 2.0813 billion tons, considering the interactions between the digital economy, data center power, and e-waste. Based on the findings, we recommend fostering innovation in the digital industry, enhancing the e-waste treatment process, strategically developing digital infrastructure, and exploring effective carbon reduction strategies for the digital economy aimed at supporting China in achieving its dual-carbon goals.

Green Data Center Cooling Control via Physics-guided Safe Reinforcement Learning

Deep reinforcement learning (DRL) has shown good performance in tackling Markov decision process (MDP) problems. As DRL optimizes a long-term reward, it is a promising approach to improving the energy efficiency of data-center cooling. However, enforcement of thermal safety constraints during DRL’s state exploration is a main challenge. The widely adopted reward-shaping approach adds negative reward when the exploratory action results in unsafety. Thus, it needs to experience sufficient unsafe states before it learns how to prevent unsafety. In this article, we propose a safety-aware DRL framework for data-center cooling control. It applies offline imitation learning and online post-hoc rectification to holistically prevent thermal unsafety during online DRL. In particular, the post-hoc rectification searches for the minimum modification to the DRL-recommended action such that the rectified action will not result in unsafety. The rectification is designed based on a thermal state transition model that is fitted using historical safe operation traces and able to extrapolate the transitions to unsafe states explored by DRL. Extensive evaluation for chilled water and direct expansion-cooled data centers in two climate conditions show that our approach saves 18% to 26.6% of total data-center power compared with conventional control and reduces safety violations by 94.5% to 99% compared with reward shaping. We also extend the proposed framework to address data centers with non-uniform temperature distributions for detailed safety considerations. The evaluation shows that our approach saves 14% power usage compared with the PID control while addressing safety compliance during the training.

Artificial intelligence and edge computing for machine maintenance-review

Industrial internet of things (IIoT) has ushered us into a world where most machine parts are now embedded with sensors that collect data. This huge data reservoir has enhanced data-driven diagnostics and prognoses of machine health. With technologies like cloud or centralized computing, the data could be sent to powerful remote data centers for machine health analysis using artificial intelligence (AI) tools. However, centralized computing has its own challenges, such as privacy issues, long latency, and low availability. To overcome these problems, edge computing technology was embraced. Thus, instead of moving all the data to the remote server, the data can now transition on the edge layer where certain computations are done. Thus, access to the central server is infrequent. Although placing AI on edge devices aids in fast inference, it poses new research problems, as highlighted in this paper. Moreover, the paper discusses studies that use edge computing to develop artificial intelligence-based diagnostic and prognostic techniques for industrial machines. It highlights the locations of data preprocessing, model training, and deployment. After analysis of several works, trends of the field are outlined, and finally, future research directions are elaborated

Deep Learning Modified Reinforcement Learning with Virtual Machine Consolidation for Energy-Efficient Resource Allocation in Cloud Computing

Cloud computing has attracted significant attention because of the growing service demands of businesses that outsource computationally intensive tasks to the data center. Meanwhile, the infrastructure of a data center is comprised of hardware resources that consume a great deal of energy and release harmful levels of carbon dioxide. Cloud data centers demand massive amounts of electrical power as modern applications and organizations grow. To prevent resource waste and promote energy efficiency, virtual machines (VMs) must be dispersed over numerous physical machines (PMs) in a data center in the cloud. The actual allocation of VMs to PMs can involve more complex decision-making processes, such as considering the resource utilization, load balancing, performance requirements, and constraints of the system. Advanced techniques, like intelligent placement algorithms or dynamic resource allocation, may be employed to optimize resource utilization and achieve efficient VM distribution across multiple PMs. Cloud service suppliers aim to lower operational expenses by reducing energy consumption while offering clients competitive services. Minimizing large-scale data center power usage while maintaining the quality of service (QoS), especially for social media-based cloud computing systems, is crucial. Consolidating VMs has been highlighted as a promising method for improving resource efficiency and saving energy in data centers. This research provides deep learning augmented reinforcement learning (RL)-based energy efficient and QoS-aware virtual machine consolidation (VMC) approach to meet the difficulties. The proposed deep learning modified reinforcement learning-virtual machine consolidation (DLMRL-VMC) model can motivate both cloud providers and customers to distribute cloud infrastructure resources to achieve high CPU utilization and good energy efficiency as measured by power usage effectiveness (PUE) and data center infrastructure efficiency (DCiE). The suggested model, DLMRL-VMC, offers a VM placement approach based on resource usage and dynamic energy consumption to determine the best-matched host and VM selection strategy, Average Utilization Migration Time (AUMT). Based on AUMT, deep learning modified reinforcement learning (DLMRL) will choose a VM with a low average CPU utilization and a short migration time. The DLMRL-VMC Energy-efficient, Resource Allocation strategy is evaluated on the trace of the CloudSim VM to attain good PUE and CPU utilization.

Thermal Characterization of Subcooled Flow Boiling in a Pin-Fin Coldplate With Non-Uniform Heating

Abstract Coldplates are a crucial component in various cooling applications, such as cooling data center servers and power electronics. The unprecedented growth in electronics power density, along with the resulting ultrahigh heat fluxes, demands a transition from single-phase forced convection to two-phase flow boiling heat transfer. The majority of studies in the literature have focused on flow boiling in fin-enhanced silicon microgaps and microchannels, with only a few addressing flow boiling in millimeter-scale heat sinks. In the present study, flow boiling of HFE-7200 dielectric fluid in a millimeter-scale pin-fin coldplate is experimentally investigated under nonuniform heating conditions. Four background heaters represent the low-dissipating-power devices. On the other hand, five hotspot heaters mimic the high-heat-flux devices and generate heat fluxes ranging from 50 W/cm2 to 1000 W/cm2, corresponding to hotspot heat inputs ranging from 62.5 W to 1.25 kW, respectively. The coldplate's thermohydraulic performance is investigated for various flow rates and inlet temperature ranging from 0.5 L/min to 1.5 L/min and from 25 °C to 60 °C, respectively. A high-speed camera is utilized for a narrow field of view (FOV) flow visualization at a frame rate of 2229 fps while a digital camera is used for a wider FOV at 60 fps. Flow visualization demonstrated the transition between bubbly, slug/churn, and stratified two-phase flow regimes.

A two-stage stochastic collaborative planning approach for data centers and distribution network incorporating demand response and multivariate uncertainties

As a critical infrastructure to support the development of digital economy, the scale of data center (DC) power consumption has been growing continuously in recent years. The plug-in of a large number of DCs to the distribution network (DN) presents both opportunities and challenges for the secure, stable, and sustainable operation of the system. In order to address the planning issues arising from the integration of DCs into the DN, this paper proposes a two-stage (planning stage and operation stage) collaborative planning method for DCs and the DN while considering the source-demand uncertainty. Firstly, spatial-temporal demand response characteristic of DCs is explicitly elaborated, and based on the energy consumption model, a two-stage stochastic planning model considering conditional value at risk (CVaR) for DCs and the DN is constructed using probabilistic stochastic simulation, which obeys low-carbon and security, and satisfies the source-network-load constraints of the system. The siting and sizing for DCs and renewable generation, feeder expansion for the DN are determined to minimize the investment cost during the planning stage, while the power output for the devices considering uncertainties from load and renewable energy are optimized to minimize the operation cost during the operation stage. Secondly, the L-shaped decomposition algorithm is proposed to solve the two-stage stochastic optimization problem by reconstructing the problem into a master problem and subproblems, where the dual multiplies are calculated to generate feasibility cuts and optimality cuts to achieve iterative solving. Finally, numerical case studies on a 33-bus distribution system demonstrate that the participation of DCs in demand response can achieve a significant reduction in the investment and operation cost of the DN, and further promote the consumption of distributed renewable energy sources.

Phase Shift Regulated Resonant Switched-Capacitor-Based Intermediate Bus Converter for 48 V Data Center Power System

Phase Shift Regulated Resonant Switched-Capacitor-Based Intermediate Bus Converter for 48 V Data Center Power System

Review of Isolated DC-DC Converters for Application in Data Center Power Delivery

Review of Isolated DC-DC Converters for Application in Data Center Power Delivery

Analysis on data center power supply system based on multiple renewable power configurations and multi-objective optimization

With rapid development of data center industry, achieving low energy consumption and costs become important. How to provide an optimal configuration on renewable distributed energy systems combining data centers is challenging. This work discussed possible hybrid configurations by combining diesel, photovoltaic, wind, and battery capacity in data center. For each mode, renewable penetration (RP) and the Levelized Cost of Electricity (LCOE) were used to evaluate the performance of system. A multi-objective optimization analysis of the case is performed based on the entropy weighted-TOPSIS method, genetic algorithm, and NSGA-II aiming to obtain both a high RP and a low LCOE. Then, comparative research on optimal power configuration with conventional or renewable energy was conducted. The results show that it is recommended to take a high-rated PV power. When wind power is taken, it is only meaningful when combined with PV power system. For the hybrid renewable power system without an energy storage unit, it's easy to realize a lower LCOE compared to Diesel mode, and its realizable maximum RP is 28.31 %. When battery energy storage is taken in the hybrid system, six optimal configurations are recommended which can achieve the lowest cost of 0.21 $/kWh and the highest RP of 44.5 %.

Advantage prioritization of digital carbon footprint awareness in optimized urban mobility using fuzzy Aczel Alsina based decision making

City governments prioritize mobility in urban planning and policy. Greater mobility in a city leads to happier citizens. Although enhanced urban mobility is helpful, it comes with costs, notably in terms of climate change. Transportation systems that enable urban mobility often emit greenhouse gases. Cities must prioritize digital carbon footprint awareness. Cities may reduce the environmental impact of urban mobility while keeping its benefits by close monitoring and reducing the carbon footprint of digital technologies like transportation applications, ride-sharing platforms, and smart traffic control systems. The aim is to advantage prioritize three alternatives, namely doing nothing, upgrading and optimizing data centers and networks, and using renewable energy sources for data centers and networks to minimize the digital carbon footprint using the proposed decision making tool. This study consists of two stages. In the first stage, fuzzy Aczel-Alsina functions (fuzzy Aczel-Alsina weighted assessment — ALWAS method) based Ordinal Priority Approach (OPA) is proposed to find the weights of criteria. Secondly, fuzzy ALWAS Combined Compromise Solution (CoCoSo) model is improved to evaluate and choose the best alternative among the three alternatives. The improved ALWAS-CoCoSo model enables flexible nonlinear processing of uncertain information and simulation of different risk levels. Besides, we proposed the improved fuzzy OPA algorithm for processing uncertain and incomplete information. The case study is provided to the decision-makers to advantage prioritize the alternatives based on twelve criteria organized into four aspects, including digital carbon footprint, externalities, technical capability, and economics. The ranking results reveal that A3=2.445 is the best among the three alternative, while A1=1.705 is the worst alternative. The results show that the best way to reduce the digital carbon footprint is to use renewable energy sources to power data centers and networks (A3).

Ultra‐Low Loss SiN Edge Coupler for III‐V/SiN Hybrid Integration

AbstractSilicon photonics is becoming a competitive technology to settle the issues of power and speed in data centers and computing systems. However, the absence of an on‐chip light source restricts the wide and deep application of silicon photonics. Essentially, efficient edge coupling from a Si‐based III‐V laser to silicon photonic components remains the major obstacle for on‐chip integration. Here, two compact edge couplers with ultra‐low coupling loss and eased fabrication for the light transmission from a Si‐based III‐V laser to Si3N4 waveguide are realized. An ultra‐low laser‐to‐chip coupling loss of only 1.175 dB for the butt coupling of a Si‐based InAs QD laser is experimentally demonstrated with high alignment tolerance. The strategies presented here provide a competitive solution for realizing ultra‐efficient integration of Si‐based light sources and silicon photonic components.

Power data integrity verification method based on chameleon authentication tree algorithm and missing tendency value

Abstract The power system operation and control data are from a wide range of sources. The relevant data acquisition equipment is disturbed by the complex electromagnetic environment on the power system operation and control lines, resulting in data errors and affecting the application and analysis of data. Therefore, a power data integrity verification method based on chameleon authentication tree algorithm and missing trend value is proposed. Get 2D data from different sensors and place it in the space environment. After data conversion, convert heterogeneous data into the same structure, expand the scope of power data acquisition, and conduct power system operation and control node layout and integrity data acquisition; The chameleon authentication tree algorithm is used to deal with the heterogeneous information of the power data, and the true value of the data is determined in the heterogeneous conflict of the power data at the same site; Query the integrity data based on the power system operation and control positioning node, creatively calculate the missing trend value of power data, evaluate the importance of data integrity, obtain the priority of power data integrity verification, and complete the integrity verification of power data. The experimental results show that the optimal clustering number is 9.05, the distribution coefficient is 16.30, the absolute error of validity analysis is 2.80, all test indicators are close to the preset standard, and the trend of the validation curve is close to the trend of the set demand covariance curve. Ensuring the integrity of power data and determining the important indicators of power lines are more conducive to the safe and stable operation of the power data center.

10th Anniversary and Powering Data Centers [From the Editor

10th Anniversary and Powering Data Centers [From the Editor

Transforming Data Center Power Infrastructure: Embracing dynamic grid support and fuel cell solutions

Transforming Data Center Power Infrastructure: Embracing dynamic grid support and fuel cell solutions

VRLA battery fault prediction for data center based on random forest model and feature enhancement method

The Valve-Regulated Lead-Acid (VRLA) battery is an important part of data center power supply system. Battery failure will threaten the safe operation of the data center. How to predict the impending battery failure in advance is an urgent problem to be solved to ensure the operation safety of the data center. However, batteries in such precise data centers rarely fail. Few faulty samples cause extreme imbalance between normal samples and faulty samples. In the data center site, the battery is usually in a floating state and the battery charge-discharge cycle times are less. As a result, the obtained battery data has a single working condition. In addition, VRLA batteries are faced with the problems of limited observation information (large amount of data but low data dimension, only voltage, resistance and temperature collected directly from sensors).In this paper, a feature enhancement method is proposed by analyzing the working characteristics of VRLA batteries in the data center. This method extends the two-dimensional characteristics (voltage, resistance) of battery to nine-dimensional characteristics to solve the problem of limited observation information of VRLA battery. The problem of extreme imbalance between normal samples and faulty samples of battery is solved based on clustering undersampling method. Based on the above two methods, a VRLA battery fault classification prediction model is proposed. The nine-month operation data of 1000 VRLA battery were randomly selected from a data center and combined with the simulated fault samples to form a test set. The test results show that the F-score value of the model is increased from 54.5 % to 97.5 % after the clustering undersampling method and the feature enhancement method proposed in this paper. Compared with the VRLA battery replacement strategy recommended in IEEE STD 1188-2005 on this test set, the method can predict the impending battery failure at least 3 days in advance.

A Harris Hawk Optimisation system for energy and resource efficient virtual machine placement in cloud data centers.

Virtualisation is a major technology in cloud computing for optimising the cloud data centre's power usage. In the current scenario, most of the services are migrated to the cloud, putting more load on the cloud data centres. As a result, the data center's size expands resulting in increased energy usage. To address this problem, a resource allocation optimisation method that is both efficient and effective is necessary. The optimal utilisation of cloud infrastructure and optimisation algorithms plays a vital role. The cloud resources rely on the allocation policy of the virtual machine on cloud resources. A virtual machine placement technique, based on the Harris Hawk Optimisation (HHO) model for the cloud data centre is presented in this paper. The proposed HHO model aims to find the best place for virtual machines on suitable hosts with the least load and power consumption. PlanetLab's real-time workload traces are used for performance evaluation with existing PSO (Particle Swarm Optimisation) and PABFD (Best Fit Decreasing). The performance evaluation of the proposed method is done using power consumption, SLA, CPU utilisation, RAM utilisation, Execution time (ms) and the number of VM migrations. The performance evaluation is done using two simulation scenarios with scaling workload in scenario 1 and increasing resources for the virtual machine to study the performance in underloaded and overloaded conditions. Experimental results show that the proposed HHO algorithm improved execution time(ms) by 4%, had a 27% reduction in power consumption, a 16% reduction in SLA violation and an increase in resource utilisation by 17%. The HHO algorithm is also effective in handling dynamic and uncertain environments, making it suitable for real-world cloud infrastructures.

A Scalable Electronic-Embedded Transformer, a New Concept Toward Ultra-High-Frequency High-Power Transformer in DC–DC Converters

The demand for high-power and high-density isolated dc-dc converter is driven by the rapid development of energy storage system, data center power supplies and transportation electrification. However, designing a high-power and high-frequency transformer presents significant challenges due to the trade-off between thermal management, leakage inductance minimization, and insulation requirements. In order to overcome this trade-off between power rating and operation frequency, a scalable electronic-embedded transformer (EET) is proposed with a low voltage bridge integrated into the transformer windings. The EET addresses the challenge through its simple open-loop control and natural current sharing, enabling easy parallel connection and scaling to different power ratings. Based on this concept, a bi-directional EET-based dc transformer (EET-DCX) is proposed to solve the transformer-level paralleling and resonant point shift issues in traditional LLC-DCX designs. By employing the embedded full bridge, the EET-DCX effectively cancels out the impedance of the leakage inductance, ensuring optimal operation at any frequency. Additionally, the EET-DCX retains the inherent advantages of the LLC-DCX, such as load-independent voltage gain, simple open-loop control, full load range zero voltage switching (ZVS), and low circulating current. Leveraging these advantages, the proposed EET-DCX solution has the potential to push the boundaries of transformer performance to the MHz operation frequency range with hundreds of kilowatts power capability. To validate these merits, a 12 kW 500 kHz EET-DCX with four planar EET units was built. This article is accompanied by two videos demonstrating the dynamic load changing and frequency changing test.