Performance Tuning Research Articles

BIOS-Based Server Intelligent Optimization.

Servers are the infrastructure of enterprise applications, and improving server performance under fixed hardware resources is an important issue. Conducting performance tuning at the application layer is common, but it is not systematic and requires prior knowledge of the running application. Some works performed tuning by dynamically adjusting the hardware prefetching configuration with a predictive model. Similarly, we design a BIOS (Basic Input/Output System)-based dynamic tuning framework for a Taishan 2280 server, including dynamic identification and static optimization. We simulate five workload scenarios (CPU-instance, etc.) with benchmark tools and perform scenario recognition dynamically with performance monitor counters (PMCs). The adjustable configurations provided by Kunpeng processing reach . Therefore, we propose a joint BIOS optimization algorithm using a deep -network. Configuration optimization is modeled as a Markov decision process starting from a feasible solution and optimizing gradually. To improve the continuous optimization capabilities, the neighborhood search method of state machine control is added. To assess its performance, we compare our algorithm with the genetic algorithm and particle swarm optimization. Our algorithm shows that it can also improve performance up to 1.10× compared to experience configuration and perform better in reducing the probability of server downtime. The dynamic tuning framework in this paper is extensible, can be trained to adapt to different scenarios, and is more suitable for servers with many adjustable configurations. Compared with the heuristic intelligent search algorithm, the proposed joint BIOS optimization algorithm can generate fewer infeasible solutions and is not easily disturbed by initialization.

Optimal Energy Shaping via Neural Approximators

We introduce optimal energy shaping as an enhancement of classical passivity-based control methods. A promising feature of passivity theory, alongside stability, has traditionally been claimed to be intuitive performance tuning along the execution of a given task. However, a systematic approach to adjust performance within a passive control framework has yet to be developed, as each method relies on few and problem-specific practical insights. Here, we cast the classic energy-shaping control design process in an optimal control framework; once a task-dependent performance metric is defined, an optimal solution is systematically obtained through an iterative procedure relying on neural networks and gradient-based optimization. The proposed method is validated on state-regulation tasks.

Digitalization of Hindu Temples in India in Google Cloud and SerpAPI Automation in Python

The study on scraping the Google Search Engine via the SerpAPI and automatically generating a dataset of the temple name, location of the temple, temple description, the state of India, longitude and latitude coordinates, and distance from the major cities in India is demonstrated. The workload orchestration is controlled by Google Composer in Apache Airflow. Google Kubernetes Engine (GKE) categorizes applications into microservices under each container. The docker image is created in the GKE control plane, API scheduler segments the granular levels of microservices in the DevOps operation of GitHub, Jenkins. YAML provides the configuration of the pod, of GKE carrying multiple containers, of the main type of input for Kubernetes configurations. Python application creates the YAML for GKE configuration data to define the Kubernetes object and the business rule. The Google native artificial intelligence VertexAPI picks the keyword of the temple name and the location of every state in India. The SerpAPI application digitized the Hindu temples in India for the complete project life cycle for requirement gathering, design, development, and deployment in Google Cloud. The search parameters are designed for Google Search Engine for the collection of historians, and architectural details of the Hindu temples in India. The Google Search application uses the Google Compute Engine, Big Query, Google Kubernetes Engine, Cloud Composer, and big data services of Data Fusion, Dataproc, and Dataflow. The solution uses Google cloud services of Web Hosting, Load balancing, Google Storage, and the compute engine performance tuning in Big Query optimization. The Google search results are in JSON format of every state in temple details and a python application parses the facts on the Search API algorithm. The temple images are visualized in Python Application and integrated for the data visualization of Google inbuilt Google Studio.

Machine Learning Guided Prediction of Superhard Materials Based on Compositional Features

In this study, the mechanical properties of materials were predicted using machine learning to search for superhard materials. Based on an AFOW database consisting of DFT quantum calculation values, the mechanical properties of materials were predicted using various machine learning models. For supervised learning, the entire data was divided into training data and test data at a ratio of 8:2. Since the discovery of superhard materials can be predicted based on the bulk modulus and shear modulus, the bulk modulus was primarily predicted using only the chemical compositional ratio (chemical formula), and then the shear modulus was obtained using the predicted bulk moduli. To obtain good prediction performance, cross-validation and hyper-parameter tuning were carried out. Each characteristic was predicted using XGBoost, one of the ensemble algorithms, and its performance was compared to the treebased machine learning of RandomForest and Support Vector Machine regression using the coefficient of determination (R<sup>2</sup>) and root-mean-square-error (RMSE) as metrics. For the recently introduced four superhard materials (Mo0.9W1.1BC, ReWC0.8, MoWC2, and ReWB), the results of this study were similar to those of previous studies including the experimental values or the DFT quantum calculations. The shear modulus was underpredicted, which can be understood since structural properties were not considering as a feature in our machine learning models.

FCNNLib: A Flexible Convolution Algorithm Library for Deep Learning on FPGAs

Convolution features huge complexity and demands high computation capability. Among hardware platforms, field programmable gate array (FPGA) emerges as a promising solution for its substantial available parallelism and energy efficiency. Besides, convolution can be implemented with different algorithms, including conventional, general matrix–matrix multiplication (GEMM), Winograd, and fast Fourier transformation (FFT) algorithms, which are diverse in arithmetic complexity, resource requirement, etc. Different convolutional neural network (CNN) models have different topologies and structures, favoring different convolution algorithms. In response, software libraries such as cuDNN provide a variety of computational primitives to support these algorithms. However, supporting such libraries on FPGAs is challenging. First, multiple algorithms can share the FPGA resources spatially as well as temporally, introducing either reconfiguration overhead or resource underutilization. Second, FPGA implementation remains a significant challenge for library developers. It typically requires significant specialized hardware knowledge. In this article, we propose <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FCNNLib</monospace> , an efficient and scalable convolution algorithm library on FPGAs. To coordinate multiple convolution algorithms on FPGAs, we develop three schedulings: 1) spatial; 2) temporal; and 3) hybrid, which exhibit different tradeoffs in latency and throughput. We explore these schedulings by balancing the reconfiguration overhead, resource utilization, and optimization objectives of the CNNs. Then, we provide efficient and tunable algorithm templates that allow performance tuning through performance and resource models. To arm the users, <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FCNNLib</monospace> exposes a set of interfaces to support high-level application designs. We demonstrate the usability of <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FCNNLib</monospace> with state-of-the-art CNNs. <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FCNNLib</monospace> achieves up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$44.6\times $ </tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.76\times $ </tex-math></inline-formula> energy efficiency in various scenarios compared with software libraries for CPUs and GPUs, respectively.

Accelerating the Machine Learning Model Deployment using MLOps

Machine Learning projects acquire a large amount of data to make predictions on the data. Deploying machine learning models is a difficult task as it is involved a lot factors such as continuous builds to improve efficiency and different libraries used for prediction. As the data grow for better predictions with proper accuracy some of the parameters are to be dynamically changed for the sake of performance tuning. To enhance the variation in the parameters it takes effort and time from the developers to make the changes in the code and deploy to the production environment. In proposed work, an end to end automation cycle to enhance the timely deployment of machine learning models with better performance is presented. This automation cycle converts the changed code with varied parameters into the container image and uploads to the container registry. Later, when the new image is uploaded it auto detects and updates the production environment with the new image data by launching a new container and shutting down the previous old one. The proposed automated model will display the updated features which are developed. The users can utilize the updated model simultaneously without any downtime to obtain the new features. The main contribution of proposed work is that it can build and deploy the machine learning models automatically, without any manual intervention. The models can also be retrained for better performance of the predictions.

Evaluation of Electrochemical Performance Tuning By Laser Structuring of Electrodes and Its Impact on Cell Degradation Mechanisms

Tuning of electrochemical properties of lithium-ion batteries by using ultrafast laser structuring of electrodes is a rather new technical approach. At Karlsruhe Institute of Technology (KIT) this technology was developed and established for the first time. However, quite recently, other research groups and institutions worldwide also recognized the high potential of this technology for lithium-ion battery production. Lately, a roll-to-roll laser processing infrastructure for structuring large footprint electrode materials for high capacity pouch cells and cylindrical cells was installed at KIT in order to push this technology to demonstrator level, namely technology readiness level 6. The high rate capability of the produced high energy battery cells were significantly improved and twice lifetime of cells could be achieved in comparison to cells with unstructured electrodes. In frame of research, cooperative, and industrial projects the impact of laser structuring for small and large footprint electrodes was studied with regard to lithium distribution caused electrochemical cycling. Hereby, different types of LiNixMnyCo1-x-yO2 (NMC) as a cathode material and graphite as well as silicon-graphite as an anode material with areal capacities up to 4 mAh/cm2 were investigated. Quantitative lithium distribution along entire electrodes were measured by laser-induced breakdown spectroscopy (LIBS). From the 3D elemental mapping starting points for electrochemical degradation could be identified. In cathode materials the local increase of lithium indicated the formation of electrical short cuts which were most related to electrode material inhomogeneity, e.g., by a macroscopic change in porosity or macroscopic film defects. The compressive stress applied to electrodes and separator during cycling has also a significant influence on lithium distribution and subsequent cell degradation. The appropriate type of laser generated patterns such as holes, lines or grids strongly depends on the application scenario as well as type of materials. For silicon-graphite anodes the huge volume expansion during lithium silicide needs to be considered and more rectangular and broader structure features become relevant. Laser structuring of electrodes showed in all cases, for small and large footprint cathodes and anodes, a rather homogenized lithium distribution along the surface and electrode thickness. For full cells elemental mapping of electrodes facing each other was performed indicating the mutual influence on lithium distribution. It can be concluded that the laser-based 3D electrode concept is beneficial regarding a reduced cell degradation. In addition, it could be proven by LIBS that for the 3D battery concept new lithium diffusion pathways were generated which becomes activated at elevated C-rates. Those new lithium diffusion pathways counteract the increase of diffusion overpotential leading to a higher capacities also for high power operation.

Stability Analysis of Coupled-Inductor-Based Buck Converter: Participation Factors Approach

A coupled-inductor-based fourth-order buck converter is introduced in this article. The salient features of this topology are lower source current ripples like in conventional buck converter with input filter, no need of connecting an additional damping network across the input filter capacitor, additional flexibility provided by the coupling in the performance tuning, and the existing input <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> – <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">_m</i> structure itself acts as a ripple steering network. The detailed steady-state analysis is evolved and conditions for source current ripple reductions are formulated. Subsequently, small-signal analysis is established and a proportional plus integral plus derivative type controller is designed to ensure load voltage regulation. To enhance the dynamic performance, a two-loop voltage-mode controller is designed. The stability analysis of the closed-loop coupled-inductor-based buck converter is carried out through the eigenvalue sensitivity and participation factors approach. The impact of converter parameters on the converter system stability is discussed. In comparison with the single-loop voltage-mode controller, the two-loop voltage-mode controller is able to ensure that the closed-loop poles stay away from the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">jω</i> -axis and, thus, the two-loop voltage-mode controller exhibits better dynamic performance. Analytical findings are validated on a 75 W, 60 to 36 V prototype converter. Steady-state and dynamic performance characterizing measurement results are given for validation of the concept.

Jack, The Autotuner

Among his many technical contributions, automated performance tuning—or autotuning—has been a critical pillar of the research that has emerged from Jack Dongarra's group at the University of Tennessee. This article reflects on the history of that work, including an analysis of the large body of research that has used or cited the ATLAS system since its inception in the late 1990 s.

Elasticsearch as a NoSQL Database for Fast Distributed Financial Service Applications

This paper explores the implementation of Elasticsearch as a primary data store for fast, distributed financial service applications. It examines the advantages of Elasticsearch in the context of financial services, including its scalability, high performance, flexible schema, and powerful search and analytics capabilities. The study addresses key challenges encountered during implementation, such as eventual consistency, limited join capabilities, and field limits, providing practical solutions based on real-world experience. The paper also discusses implementation strategies, focusing on data modeling for complex financial instruments, query optimization, and performance tuning. By balancing Elasticsearch's strengths against its limitations, this case study demonstrates how financial service organizations can leverage this technology to build more scalable, performant, and adaptable systems capable of handling the increasing data volumes and real-time processing demands of modern finance.

Real-Time Performance Analysis and Tuning of Embedded System Virtualization Architecture Based on KVM

Embedded systems are proceeding towards exploiting virtualization technology to have the benefits of Real-Time Operating System (RTOS) and General-Purpose Operating System (GPOS) in the same system. This combination provides both a timely and deterministic behavior and a general-purpose application codebase. There still exist concerns about the real-time responsiveness of RTOS running inside a Virtual Machine (VM). In this paper, the real-time performance of Kernel-based Virtual Machine (KVM) virtualization architecture is analyzed on a multi-core system. Here, a preemptible Linux kernel with the PREEMPT_RT patch is used for RTOS, while a standard Linux kernel is used for GPOS. The interrupt latency inside the real-time guest VM is analyzed by applying various amounts of CPU, memory, and I/O stresses on the guest and host systems. A VM resource monitoring tool ‘VM_stat’ is developed to know the resource usage of the guest VMs, which is useful for effectively tuning the system. Different real-time tuning measures are applied on the host/guest systems and the performance is analyzed.

FSbrain: An intelligent I/O performance tuning system

To bridge the performance gap between network and storage, performance tuning of file systems according to different workloads has become an important work on the Internet of Things (IoT). However, existing methods on file system tuning mainly focused on tailoring the file system kernel codes, or cannot scope to high dimensional parameters and dynamically changing workloads. In this paper, we propose an automatic I/O performance tuning system FSbrain that recommends reasonable and performance-efficient configurations for file systems by using machine learning. It makes the configuration recommendations based on current workloads within an acceptable time, however, without any manual interventions or kernel code modifications. In specific, FSbrain mainly consists of two phases which are model training and configuration tuning. Since tuning attempts will cause frequent remounting on target file systems, we deploy FSbrain and conduct the tuning sessions on an experimental node called “shadow server”. We evaluate FSbrain on three representative file systems (namely, Ext4, F2FS, and PMFS). The experimental results show that the configuration recommended by FSbrain can improve the I/O performance by up to 1.28× averagely than the default configuration. Besides, FSbrain reduces the overall tuning time by 90% compared to manual tuning.

ERP system usage effect on organisational performance: The impact of performance tuning and post implementation changes

With an emphasis on performance tuning and post-implementation modifications, this study examines how ERP system usage affects organisational performance. Modern company operations cannot function without Enterprise Resource Planning (ERP) solutions, which provide improved data management capabilities and expedite procedures. The amount of usage, optimisation through performance tuning, and flexibility to post-implementation changes are some of the elements that determine their influence on organisational performance. This study delves into the link between ERP system usage and organisational performance, specifically looking at how performance tuning and post-implementation adjustments impact efficiency, productivity, and competitiveness. It does this through a thorough literature review and empirical analysis. The results shed light on the dynamics of ERP systems and provide useful information for practitioners who want to get the most out of ERP projects.

Laser‐Based Growth and Treatment of Graphene for Advanced Photo‐ and Electro‐Related Device Applications

AbstractGraphene is considered as a competitive material for the development of photo‐ and electrodevices because of its superior electrical and optical properties. Meanwhile, laser processing technologies with high precision, fast reaction speed, and high maneuverability are widely used for the fast and facile fabrication of functionalized graphene and relevant devices. In particular, it enables desired performance tuning, programmable structural designing, micro‐ or nanoscale processing, and large‐scale integration for advanced graphene‐based photo‐ and electro‐related devices, which has greatly promoted the development of multiscale and multidimensional device manufacturing technology. In this review, the interaction principle and processing strategies of functional graphene prepared and treated by laser, especially for the regulation of optical and electrical properties, will be fully understood and introduced. The latest progress and reports of advanced photoelectro devices, including electric energy storage devices, generators, sensors, actuators, photodetectors, photothermal devices, and photovoltaic devices, are classified and summarized. In general, this review may provide positive and meaningful guidance for flourishing graphene‐based photo‐ and electroindustries.

Predicting number of threads using balanced datasets for openMP regions

Incorporating machine learning into automatic performance analysis and tuning tools is a promising path to tackle the increasing heterogeneity of current HPC applications. However, this introduces the need for generating balanced datasets of parallel applications’ executions and for dealing with natural imbalances for optimizing performance parameters. This work proposes a holistic approach that integrates a methodology for building balanced datasets of OpenMP code-region patterns and a way to use such datasets for tuning performance parameters. The methodology uses hardware performance counters to characterize the execution of a given region and correlation analysis to determine whether it covers an unique part of the pattern input space. Nevertheless, a balanced dataset of region patterns may become naturally imbalanced when used for training a model for tuning any specific performance parameter. For this reason, we have explored several methods for dealing with naturally imbalanced datasets for finding the appropriated way of using them for tuning purposes. Experimentation shows that the proposed methodology can be used to build balanced datasets and that such datasets, plus a combination of Random Forest and binary classification, can be used to train a model able to accurately tune the number of threads of OpenMP parallel regions.

Compression Induced Deformation Twinning Evolution in Liquid-Like Cu2Se.

For practical applications of copper selenide (Cu2Se) thermoelectric (TE) materials with liquid-like behavior, it is essential to determine the structure-property relations as a function of temperature. Here, we investigate β-Cu2Se structure evolution during uniaxial compression over the temperature range of 400-1000 K using molecular dynamics simulations. We find that at temperatures above 800 K, Cu2Se exhibits poor stability with breaking order that is described as a liquid-like or hybrid structure comprising a rigid Se sublattice and mobile Cu ions. A uniaxial load causes accumulated structural heterogeneity that is alleviated by diffusion-induced accommodation of local deformations. With increasing strain, the deformation mode changes into a combination of compression and shear, accompanied by restructuring in terms of twinning. Interestingly, in addition to a plastic behavior rarely found in inorganic semiconductors, we find that higher temperature promotes deformation twinning in liquid-like Cu2Se, showing the role of thermal instability, including Cu diffusion, in structural adaptation and mechanical modulation. These findings reveal the micromechanism of hybrid structural evolution as well as performance tuning through twinning, which provides a theoretical guide toward advanced Cu2Se TE materials design.

Non-Singular Finite Time Tracking Control Approach Based on Disturbance Observers for Perturbed Quadrotor Unmanned Aerial Vehicles.

In this paper, a disturbance observer based on the non-singular terminal sliding mode control method was presented for the quadrotor in the presence of wind perturbation. First, the position and attitude dynamical equation of the quadrotor was introduced in the existence of windy perturbation. It was difficult to exactly determine the upper bound of the perturbations in the practical systems such as robot manipulators and quadrotor UAVs. Then, a disturbance observer was designed for the estimation of wind perturbation which was entered to the quadrotor system at any moment. Afterward, a non-singular terminal sliding surface was proposed based on the disturbance observer variable. Furthermore, finite time convergence of the closed-loop position and attitude models of the quadrotor was proved using Lyapunov theory concept. Unlike the existing methods, the new adaptive non-singular terminal sliding mode tracker for quadrotor unmanned aerial vehicles enabled accurate tracking control, robust performance, and parameter tuning. Through the combination of the finite time tracker and disturbance observer, the position and attitude tracking of quadrotor UAVs could be accurately performed not only in the nominal environment but also in the existence of different types of perturbations. Finally, simulation results based on the recommended method were provided to validate the proficiency of the suggested method. Moreover, comparison results with another existing study were presented to prove the success of the proposed method.

Statistical analysis of an improved tuning method for optimizing performances of Hadoop applications

Hadoop is one among upcoming and most promising technologies for handling immense volume of data, generated from a variety of sources. It supports parallel processing among multiple jobs, and when such jobs are being executed on Hadoop, it is of paramount importance to know if Hadoop cluster resources are being completely utilized or not, which in turn tells us about efficiency in resource utilization for the purpose of effective completion of jobs. It could happen that Hadoop cluster resources are not being completely utilized, and because of which there arises a need for performance tuning. Fortunately, Hadoop framework equips us with several parameters to do the same and fine-tune aforesaid jobs. Performance tuning comprises of four main components: CPU utilization, Memory consumption, Disk I/O and Network traffic. In the proposed work, authors have presented several important relative parameters in relation to aforesaid components, along with how to fine-tune these parameters with appropriate values so as to optimize Hadoop execution.

A comparative performance study of random‐grid model for hyperparameters selection in detection of abnormalities in digital breast images

AbstractDeep learning models have been widely reported to have achieved significant performance in image processing and classification tasks. They have mainly been harnessed and applied to the problem of detecting abnormalities in digital breast images. However, the significant number and high dimensional space requirement of hyperparameters in deep learning models often make it challenging to find the best configuration for such parameters when tuning for best performance. In appropriately handling this performance tuning often lead to difficulty in striking balance between underfitting and overfitting. This article proposes an optimized convolutional neural network (CNN) architecture through application of hybrid selection model in obtaining best hyperparameter configuration which outperforms similar existing models. We approached this non‐trivial challenge by defining a hybrid of the grid‐based and random‐based model for the selection of hyperparameters and then investigate the performance of the configurations. To further improve the performance of the CNN model, data augmentation technique was applied. Furthermore, the study undertook a comparative study of the performance of the best configuration on some benchmarked datasets. The resulting model was applied to publicly available benchmark datasets, namely, the DDSM and MIAS datasets. Findings from the experimentations revealed that hyperparameters with Adam optimization algorithm showed superiority by yielding an accuracy of 1.0 using DDSM dataset, while SGD, RMSprop, Adam, and Adagrad output an accuracy of 0.9375 with MIAS dataset. The outcome of this study further strengthens the appropriateness of Adam optimizer and has also produced a state‐of‐the‐art CNN model suitable for solving the problem of detection and classification of breast cancer from digital mammography.

Pegasus: Performance Engineering for Software Applications Targeting HPC Systems

Developing and optimizing software applications for high performance and energy efficiency is a very challenging task, even when considering a single target machine. For instance, optimizing for multicore-based computing systems requires in-depth knowledge about programming languages, application programming interfaces, compilers, performance tuning tools, and computer architecture and organization. Many of the tasks of performance engineering methodologies require manual efforts and the use of different tools not always part of an integrated toolchain. This paper presents Pegasus, a performance engineering approach supported by a framework that consists of a source-to-source compiler, controlled and guided by strategies programmed in a Domain-Specific Language, and an autotuner. Pegasus is a holistic and versatile approach spanning various decision layers composing the software stack, and exploiting the system capabilities and workloads effectively through the use of runtime autotuning. The Pegasus approach helps developers by automating tasks regarding the efficient implementation of software applications in multicore computing systems. These tasks focus on application analysis, profiling, code transformations, and the integration of runtime autotuning. Pegasus allows developers to program their strategies or to automatically apply existing strategies to software applications in order to ensure the compliance of non-functional requirements, such as performance and energy efficiency. We show how to apply Pegasus and demonstrate its applicability and effectiveness in a complex case study, which includes tasks from a smart navigation system.