Data Center Applications Research Articles

Experimental Study on the Performance of a Novel Data Center Air Conditioner Combining Air Cooling and Evaporative Cooling

Addressing challenges of high-temperature shutdowns, low energy efficiency, and limited natural cooling utilization in data center air conditioning, a novel hybrid air cooling and evaporative cooling air-conditioning(ACEC-AC) system has been developed. The system is well-suited for application in data centers located in regions with high outdoor temperatures and large dry-bulb to wet-bulb temperature differences during summer. Experimental evaluations in an enthalpy difference laboratory revealed that the hybrid system's performance in wet mode (combining air cooling and evaporative cooling) significantly outperforms its dry mode (solely air cooling) and traditional cooling units. At an outdoor temperature of 35.60°C dry-bulb and 29.95°C wet-bulb, the system achieved a COP of 3.85, a 21.45% improvement over dry mode and 16.16% higher than conventional Packaged computer room air conditioning (CRAC). Notably, the system maximizes its benefits in regions like Guangzhou, where the temperature difference between the dry bulb and wet bulb is significant, resulting in an annual operational cost savings of over 34.16%. In conclusion, the hybrid air cooling and evaporative cooling system represents an effective energy-saving technology with great potential to enhance data center performance and reduce energy consumption.

Mixed domain coherent link with electrically reconfigurable IMDD and coherent modes for O-band data center applications

We report the first O-band link with electrically reconfigurable intensity-modulation direct-detection (IMDD) and coherent operation using custom silicon photonic chips packaged with commercial electronic chips. Transmission below the KP4-FEC threshold is shown using commercial 53 Gbaud PAM4 digital signal processing (DSP) for 16QAM (200 Gbps/λ) and PAM4 (100 Gbps/λ). Efficient operation of the packaged full link at 12 and 11.2 pJ/bit is achieved for the PAM4 and 16QAM modes, respectively.

Numerical simulation of a hybrid energy system proposed for low carbon data center

Low-carbon energy application for both large grid and distributed energy consumers is an urgent problem to be solved. Data centers (DC) are the typical distributed large-size energy consumers, and the application of renewable energy and energy storage is a promising solution for data center decarbonization. This paper developed a complete hybrid energy system (HES) simulation process for data center applications. The photovoltaic (PV) cell, hydrogen electrolysis, hydrogen storage, proton exchange membrane fuel cell (PEMFC), lithium battery (BAT), grid, and data center load are considered in the HES. A one-dimensional two-phase flow PEMFC model is developed. The detailed mass transfer process inside the PEMFC cell is modeled. The PEMFC model-solving algorithm is developed and integrated into the HES simulation program. The energy management system (EMS) based on a logical flow chart energy management strategy is integrated by editable subfunction. The HES simulation program is coded and implemented based on the MATLAB-SIMULINK environment. The yearly HES performance simulations are conducted and analyzed for a 200 kW data center in Xi’an with different HES parameters and an appropriate HES parameter arrangement is provided. The main contribution of this work is the development of an editable HES simulation program that is extensible for data center applications.

A Switched-Capacitor and Series-Resonant Hybrid MHz DCX in Data Center Applications

A Switched-Capacitor and Series-Resonant Hybrid MHz DCX in Data Center Applications

Hydrogen as power storage technology, polymeric and interconnect material innovations for future AI datacenter applications: a review

Hydrogen as power storage technology, polymeric and interconnect material innovations for future AI datacenter applications: a review

An Improved Non-Isolated Resonant Converter With Integrated Magnetics for Data Center Applications

An Improved Non-Isolated Resonant Converter With Integrated Magnetics for Data Center Applications

A fuel cell powered rack prototype for data center application with waste heat recovery: design consideration and experimental verification

The combination of data center and PEMFC has been a promising data center zero-carbon emission solution. However, the relevant experimental studies are very rare and most work only solved the simple electrical coupling. This paper proposed a new in-rack data center servers and PEMFC coupling design scheme considering both the electrical coupling and thermal management coupling of the servers and PEMFC, and built a cabinet prototype for the first time. The complete multi-condition continuous experiment on the PEMFC system fully demonstrates the servers power supply stability of the prototype with the fast tested power response capability of 4.9 kW in 0.1 s. The experiments under rated condition (when the servers power is 5 kW) were conducted and the carbon emission level and efficiency performance of the prototype are analyzed. Under the rated condition, the highest prototype CHP efficiency is 64.1 % while the prototype can provide 41.6 ℃ hot water with 3.04 L·min-1 and the lowest prototype CHP efficiency is 59.8 % while the prototype can provide 61.1 ℃ hot water with 1.33 L·min-1. The prototype maintains the zero-carbon (or negative carbon) emission of servers. The average operation carbon emission rate is -35 g·h-1. Compared with the full grid supporting data center, the power average prototype operation carbon reduction rate is 0.13 g·s-1·kW-1. The application potential of the proposed data center and PEMFC coupling scheme is demonstrated with the prototype experiment results from the advantage of carbon emission and heat recovery potential.

Low‐Defect Quantum Dot Lasers Directly Grown on Silicon Exhibiting Low Threshold Current and High Output Power at Elevated Temperatures

The direct growth of III‐V materials on silicon is a key enabler for developing monolithically integrated lasers, offering substantial potential for ultradense photonic integration in vital communications and computing technologies. However, the III‐V/Si lattice and thermal expansion mismatch pose significant hurdles, leading to defects that degrade lasing performance. This study overcomes this challenge, demonstrating InAs/GaAs‐on‐Si lasers that perform on par with top‐tier lasers on native GaAs substrates. This is achieved through a newly developed epitaxial approach comprising a series of rigorously optimized growth strategies. Atomic‐resolution scanning tunneling microscopy and spectroscopy experiments reveal exceptional material quality in the active region and elucidate the impact of each growth strategy on defect dynamics. The optimized III‐V‐on‐silicon ridge‐waveguide lasers demonstrate a continuous‐wave threshold current as low as 6 mA and high‐temperature operation reaching 165 °C. At 80 °C, critical for data center applications, they maintain a 12 mA threshold and 35 mW output power. Furthermore, lasers fabricated on both Si and GaAs substrates using identical processes exhibit virtually identical average threshold current. By eliminating the performance limitations associated with the GaAs/Si mismatch, this study paves the way for robust and high‐density integration of a broad spectrum of critical III‐V photonic technologies into the silicon ecosystem.

Optimizing a dew point evaporative cooler in data center applications

Abstract The escalating energy consumption of data centers has led to a pressing need for energy-efficient cooling solutions. This paper presents a countercurrent dew point evaporative cooler (DPEC) for data center refrigeration. We developed and experimentally validated a numerical model for DPEC, then formulated regression models using the response surface method. These models link eight key design factors, including geometrical and operational factors, to three performance indices: cooling capacity per unit volume, coefficient of performance, and outlet primary air temperature. We assessed the extent of factor influence on these indices. By using these regression models as objective functions, we used the genetic algorithm for design optimization under two climatic conditions, resulting in various optimal parameter combinations. Our findings highlight the strong predictive accuracy of these models. In comparison to the original design, the optimal design achieved an improvement of 104.8%, an increase of 23.9%, and a reduction of 13.8% in the three indices.

Recent Advances in Low-Power Digital Signal Processing Technologies for Data Center Applications

Recent Advances in Low-Power Digital Signal Processing Technologies for Data Center Applications

Self-Frequency Tracking for Fixed-Ratio Switch-Capacitors in Data Center Application without Extra Sensors

As research into data centers progresses, the importance of resonant switched-capacitor converters in power supply design becomes more evident. Practical applications reveal that the values of resonant capacitors and inductors may deviate from their nominal values due to various factors, leading to resonant frequency instability. This instability poses a challenge to power electronics technology, affecting system reliability and performance. This paper analyzes the effect of frequency deviation on system functionality, identifies the relationship between output voltage and switching frequency, and proposes a self-tracking frequency strategy to address this issue. Through experimental validation, this approach shows that it maintains synchronization between switching and resonant frequencies, reducing losses associated with frequency misalignment. Simulation and experimental results validate the converter’s stable operation and its ability to achieve zero-voltage and zero-current switching.

Compiler-Driven FPGA Virtualization with SYNERGY

FPGAs are increasingly common in modern applications, and cloud providers now support on-demand FPGA acceleration in datacenters. Applications in datacenters run on virtual infrastructure, where consolidation, multi-tenancy, and workload migration enable economies of scale that are fundamental to the provider’s business. However, a general strategy for virtualizing FPGAs has yet to emerge. While manufacturers struggle with hardware-based approaches, we propose a compiler/runtime-based solution called Synergy . We show a compiler transformation for Verilog programs that produces code able to yield control to software at sub-clock-tick granularity according to the semantics of the original program. Synergy uses this property to efficiently support core virtualization primitives: suspend and resume, program migration, and spatial/temporal multiplexing, on hardware which is available today . We use Synergy to virtualize FPGA workloads across a cluster of Intel SoCs and Xilinx FPGAs on Amazon F1. The workloads require no modification, run within 3–4 x of unvirtualized performance, and incur a modest increase in FPGA fabric usage.

Silicon-Nanowire-Based 100-GHz-Spaced 16λ DWDM, 800-GHz-Spaced 8λ LR-8, and 20-nm-Spaced 4λ CWDM Optical Demultiplexers for High-Density Interconnects in Datacenters

Several types of silicon-nanowire-based optical demultiplexers (DeMUXs) for use in short-reach targeted datacenter applications were proposed and their spectral responses were experimentally verified. First, a novel 100-GHz-spaced 16λ polarization-diversified optical DeMUX consisting of 2λ delayed interferometer (DI) type interleaver and 8λ arrayed waveguide gratings will be discussed in the spectral regimes of C-band, together with experimental characterizations showing static and dynamic spectral properties. Second, a novel 800-GHz-spaced 8λ optical DeMUX was targeted for use in LR (long reach) 400 Gbps Ethernet applications. Based on multiple cascade-connected DIs, by integrating the extra band elimination cutting area, discontinuous filtering response was analytically identified with a flat-topped spectral window and a low spectral noise of <−20 dB within an entire LR-8 operating wavelength range. Finally, a 20-nm-spaced 4λ coarse wavelength division multiplexing (CWDM)-targeted optical DeMUX based on polarization diversity was experimentally verified. The measurement results showed a low excessive loss of 1.0 dB and a polarization-dependent loss of 1.0 dB, prominently reducing spectral noises from neighboring channels by less than −15 dB. Moreover, TM-mode elimination filters were theoretically analyzed and experimentally confirmed to minimize unwanted TM-mode-oriented polarization noises that were generated from the polarization-handling device. The TM-mode elimination filters functioned to reduce polarization noises to much lower than −20 dB across the entire CWDM operating window.

Deep learning algorithms for temperature prediction in two-phase immersion-cooled data centres

PurposeAs data centres grow in size and complexity, traditional air-cooling methods are becoming less effective and more expensive. Immersion cooling, where servers are submerged in a dielectric fluid, has emerged as a promising alternative. Ensuring reliable operations in data centre applications requires the development of an effective control framework for immersion cooling systems, which necessitates the prediction of server temperature. While deep learning-based temperature prediction models have shown effectiveness, further enhancement is needed to improve their prediction accuracy. This study aims to develop a temperature prediction model using Long Short-Term Memory (LSTM) Networks based on recursive encoder-decoder architecture.Design/methodology/approachThis paper explores the use of deep learning algorithms to predict the temperature of a heater in a two-phase immersion-cooled system using NOVEC 7100. The performance of recursive-long short-term memory-encoder-decoder (R-LSTM-ED), recursive-convolutional neural network-LSTM (R-CNN-LSTM) and R-LSTM approaches are compared using mean absolute error, root mean square error, mean absolute percentage error and coefficient of determination (R2) as performance metrics. The impact of window size, sampling period and noise within training data on the performance of the model is investigated.FindingsThe R-LSTM-ED consistently outperforms the R-LSTM model by 6%, 15.8% and 12.5%, and R-CNN-LSTM model by 4%, 11% and 12.3% in all forecast ranges of 10, 30 and 60 s, respectively, averaged across all the workloads considered in the study. The optimum sampling period based on the study is found to be 2 s and the window size to be 60 s. The performance of the model deteriorates significantly as the noise level reaches 10%.Research limitations/implicationsThe proposed models are currently trained on data collected from an experimental setup simulating data centre loads. Future research should seek to extend the applicability of the models by incorporating time series data from immersion-cooled servers.Originality/valueThe proposed multivariate-recursive-prediction models are trained and tested by using real Data Centre workload traces applied to the immersion-cooled system developed in the laboratory.

Instant queue occupancy used for automatic traffic scheduling in data center networks

Datacenter applications desire low latency for short messages to provide a better user experience. Therefore, one of the goals of datacenter networks is to minimize flow completion time (FCT), especially for short flows. Multiple scheduling disciplines have been proposed to achieve this goal. In this paper, we develop a Dynamic Longer Stay Less Priority (D-LSLP) which looks at the current queue occupancy to adjust the demotion threshold for the packets of the flows arriving in the strict priority queues. Initially, D-LSLP considers every flow to be a short flow, and with the passage of time, the flow is demoted to the lower priority queue, similar to Multilevel Feedback Queue (MLFQ) Scheduling. It enables short flows to be completed in a couple of higher-priority queues. In contrast, large flows are demoted to the lower priority queues after remaining active in the queues for a certain amount of time (demotion threshold). However, for the different traffic patterns with different distribution of flow lengths, the demotion threshold for the flows should be adapted automatically. In this paper, we leverage the programmable nature of the P4 switches to measure the instant queue occupancy in the P4, which can be used to adjust the demotion threshold accordingly for the highest priority queue. This enables D-LSLP to increase or decrease its highest priority demotion threshold based on the instant queue status. D-LSLP allows multiple traffic patterns to coexist without manually tuning the demotion thresholds. Furthermore, it reduces the tail drop in the highest priority queue when many short flows overwhelm the highest priority queue. The performance evaluation shows that it works equally well for different traffic patterns without operator intervention.

A methodological framework for optimizing the energy consumption of deep neural networks: a case study of a cyber threat detector

AbstractThe growing prevalence of deep neural networks (DNNs) across various fields raises concerns about their increasing energy consumption, especially in large data center applications. Identifying the best combination of optimization techniques to achieve maximum energy efficiency while maintaining system performance is challenging due to the vast number of techniques available, their complex interplay, and the rigorous evaluation required to assess their impact on the model. To address this gap, we propose an open-source methodological framework for the systematic study of the influence of various optimization techniques on diverse tasks and datasets. The goal is to automate experimentation, addressing common pitfalls and inefficiencies of trial and error, saving time, and allowing fair and reliable comparisons. The methodology includes model training, automatic application of optimizations, export of the model to a production-ready format, and pre- and post-optimization energy consumption and performance evaluation at inference time using various batch sizes. As a novelty, the framework provides pre-configured "optimization strategies" for combining state-of-the-art optimization techniques that can be systematically evaluated to determine the most effective strategy based on real-time energy consumption and performance feedback throughout the model life cycle. As an additional novelty, "optimization profiles" allow the selection of the optimal strategy for a specific application, considering user preferences regarding the trade-off between energy efficiency and performance. Validated through an empirical study on a DNN-based cyber threat detector, the framework demonstrates up to 82% reduction in energy consumption during inference with minimal accuracy loss.

On the scalability of the speedup considering the overhead of consolidating virtual machines in servers for data centers

Virtualization technologies are extensively utilized in data centers, particularly cloud computing. This facilitates data center management and diminishes the number of physical machines (servers) and, subsequently, their cooling requirements, leading to cost, space, and power consumption reductions. When applications in data centers are executing independent parallel transactions, but with similar performance requirements, the appropriate level of virtual machine consolidation on a server poses a fundamental challenge for capacity planning. This article introduces a method to evaluate the performance speedup achieved through virtualization on any server and the effects of virtualization and consolidation overheads on physical or virtual machine scalability. This research formalizes the speedup and overheads, using classical computer architecture statements. but at the same time proposes a new method to analyze these overhead amounts and types, showing the scalability and efficiency of different consolidations in the same server and its comparison against no consolidation. This work also proposes a new way to determine the optimal number of physical servers and the optimal number of consolidated virtual machines for a given transaction workload. The real experimentation was performed with different workload sizes, types of virtualizations and different servers. The method presented also facilitates the representation of linear scalability against the real degree of parallelism of either physical machines or consolidated virtual machines for a given transaction workload, as well as striking the right balance between speedup and energy in virtual server consolidation.

Assessment of Low Polarization Dependent Multicast and Select Switch Based on Bulk SOA for Data Center Application

Assessment of Low Polarization Dependent Multicast and Select Switch Based on Bulk SOA for Data Center Application

Unveiling the evolution of light within photonic integrated circuits

Silicon photonics leverages mature semiconductor technology to produce cost-effective and high-performance components for various applications in data centers, artificial intelligence, and quantum computing. While the geometry of photonic integrated circuits can be characterized by existing means, their optimal and accurate performance requires detailed characterization of the light propagating within them. Here we demonstrate the first, to our knowledge, direct visualization of the light as it travels inside photonic integrated circuits. We employ the natural nonlinear optical properties of silicon to directly map the electric field of the waves guided inside the integrated circuits, characterizing waveguides and multimode splitters while extracting various parameters of the device—all in real-time and in a noninvasive manner. Our approach for visualizing light inside photonic circuits is the only solution directly providing such information without any overhead or penalty, potentially making it a crucial component for the characterization of photonic circuitry, toward their improved design, fabrication, and optimization.

Design and Analysis of Stochastic Processing and Matching Networks

Stochastic Processing Networks (SPNs) [24] are ubiquitous in engineering with applications in Data Centers (eg. packet routing), Telecommunication (eg. call centers), Transportation, etc. Stochastic Matching Networks (SMNs) are newly emerging variation of SPNs where 'jobs' and 'servers' arrive in the system and are matched using a decision rule. SMNs have applications in Online matching markets (eg. uber platform), Healthcare (eg. kidney transplant), Quantum Networks, Assemble-to-order (eg. manufacturing plants) and many other crucial problems. The aim of my research is to design provably efficient decision-making policies for Stochastic Processing Networks (SPNs) and Stochastic Matching Networks (SMNs) and develop analytical tools to obtain performance guarantees on their throughput and delay.