Performance Optimization

Cloud-based scientific computing necessitates effective computational models in order to improve
execution time, accuracy, and resource consumption. To measure computational efficiency and
scalability, this study compares the performance of Genetic Algorithms (GA), Monte Carlo
Methods (MCM), and Markov Models (MM) in a cloud context. The study formalizes the
mathematical foundations of each technique and evaluates its effectiveness using controlled
simulations. A comparison analysis and ablation study show that hybrid models increase
performance, with the Full Model obtaining the highest accuracy (94.47%) and the best resource
utilization (90.42%). The study found that combining several computational methodologies
improves efficiency, decision-making, and adaptability for large-scale scientific computations in
cloud environments.

Currently, several of the high performance processors used in a PC cluster have a DVS (Dynamic Voltage Scaling) architecture that can dynamically scale processor voltage and frequency. Adaptive scheduling of the voltage and frequency enables us to reduce power dissipation without a performance slowdown during communication and memory access. In this paper, we propose a method of profiledbased power-performance optimization by DVS scheduling in a high-performance PC cluster. We divide the program execution into several regions and select the best gear for power efficiency. Selecting the best gear is not straightforward since the overhead of DVS transition is not free. We propose an optimization algorithm to select a gear using the execution and power profile by taking the transition overhead into account. We have built and designed a power-profiling system, PowerWatch. With this system we examined the effectiveness of our optimization algorithm on two types of power-scalable clusters (Crusoe and Turion). According to the results of benchmark tests, we achieved almost 40% reduction in terms of EDP (energy-delay product) without performance impact (less than 5%) compared to results using the standard clock frequency.

In the ever-evolving digital age, distributed web systems are an important part of the online infrastructure, delivering services and applications on a global scale. However, there are growing concerns about the exponential growth of energy consumption in such systems, which has a significant impact on operating costs and the environment. The article addresses key topics such as energy efficiency as well as optimization strategies, including source code optimization, the use of energy-saving technologies, and intelligent resource management. The paper provides a thorough analysis of the principals of energy efficiency in distributed web systems and serves as a basis for further research and the creation of practical solutions for a more energyefficient online world. The purpose of this paper is to study strategies and methods for optimizing energy efficiency, with an emphasis on reducing energy consumption without compromising performance or availability.

This paper presents an approach to improving performance in distributed web systems by using Docker Swarm, a container orchestration tool built into Docker that allows for the management of a cluster of Docker engines. It is important to efficiently manage tasks and resources within a distributed environment due to the demand for scalable web services and increased availability. Docker Swarm provides a simple and efficient approach to managing Docker containers across multiple nodes, enabling the automatic distribution and scaling of web applications. In this article, we explore the benefits of using Docker Swarm in the context of distributed web systems, such as optimized resource management, scalability, and service reliability. Moreover, this text provides practical examples and case studies to demonstrate how the Docker Swarm framework can improve the performance and availability of distributed web applications in diverse production environments. Additionally, it outlines the benefits and difficulties related to using Docker Swarm and explores potential avenues for enhancing the performance of distributed web systems with this technology.

Cloud API performance optimization through machine learning ensemble cache management represents a significant advancement in distributed systems technology. The implementation of an innovative framework combining Random Forest, XGBoost, and Support Vector Machines creates a dynamic caching strategy that adapts to varying API workloads and usage patterns. This ensemble mechanism processes historical request patterns, payload characteristics, and temporal dependencies to predict cache relevancy and optimize replacement policies. The framework incorporates deep learning techniques for pattern recognition, gradient boosting for temporal analysis, and specialized kernels for classification tasks.

Since the operating mode of 1.55 µm AlN/GaN-based intersubband photodetectors is based on optical rectification, both the excited state lifetime and the lateral displacement of the carriers play an important role for performance optimization. We thus show here results of an improved detector generation based on a novel type of active region. Thanks to the use of quantum dots instead of quantum wells, a factor of 60 could be gained in terms of maximum responsivity. In addition, the maximum performance was achieved at a considerably higher temperature of 160 K instead of 80 K as typically seen for quantum wells.

Adaptive optics (AO) is essential for many elements of the science case for the Thirty Meter Telescope (TMT). The initial requirements for the observatory's facility AO system include diffraction-limited performance in the near IR, with 50 per cent sky coverage at the galactic pole. Point spread function uniformity and stability over a 30 arc sec field-ofview are also required for precision photometry and astrometry. These capabilities will be achieved via an order 60x60 multi-conjugate AO system (NFIRAOS) with two deformable mirrors, six laser guide star wavefront sensors, and three low-order, IR, natural guide star wavefront sensors within each client instrument. The associated laser guide star facility (LGSF) will employ 150W of laser power at a wavelength of 589 nm to generate the six laser guide stars. We provide an update on the progress in designing, modeling, and validating these systems and their components over the last two years. This includes work on the layouts and detailed designs of NFIRAOS and the LGSF; fabrication and test of a full-scale prototype tip/tilt stage (TTS); Conceptual Designs Studies for the real time controller (RTC) hardware and algorithms; fabrication and test of the detectors for the laser-and natural-guide star wavefront sensors; AO system modeling and performance optimization; lab tests of wavefront sensing algorithms for use with elongated laser guide stars; and high resolution LIDAR measurements of the mesospheric sodium layer. Further details may be found in specific papers on each of these topics.

The Extreme Ultraviolet Explorer mission is described. For the first six months, an all-sky survey will be carried out covering 90 to 750 A, or essentially the entire extreme ultraviolet (EUV) bandpass. This EUV survey will be made in four bands, or colors: U 90-150 A, 170-250 A, 400-600 A, and 550-750 A. A portion of the sky which is free from the normally intense 304 A geocoronal helium background will be surveyed at greater sensitivity; the wavelength coverage of this band is from 90 to 400 A. Following the sky survey portion of the mission, spectroscopy of individual sources will be carried out. Three spectrometers employing novel variable line-space gratings will provide spectra with -1 A resolution over the band from 70 to 760 A. This spectroscopy will be carried out by guest observers chosen by NASA in a manner roughly analogous to the International Ultraviolet Explorer (IUE) guest observer program.

The changing scene of edge computing is causing a revolution in technology, yet significant research gaps exist in understanding its full potential and implementation challenges. Through a systematic analysis of 78% of recent literature, combined with quantitative performance evaluations and real-world case studies, this research addresses critical gaps in architectural frameworks, standardization protocols, and implementation metrics. Using a three-phase methodology combining systematic literature review, qualitative trend analysis, and quantitative performance assessment, our research reveals significant improvements in operational efficiency. The systematic review identified key architectural patterns, while performance testing demonstrated a 65% reduction in response time, 72% less data transfer requirements, and 99.95% system uptime. Through comparative analysis of edge-deployed AI models, we validated new lightweight architectures that are 75% smaller while maintaining 95% accuracy. Our federated learning experiments established effective privacy-preserving methods, successfully balancing data utility with compliance requirements. Industry case studies, conducted through rigorous field testing and performance monitoring, validated our findings in manufacturing and healthcare sectors. In manufacturing, controlled experiments achieved 99.3% accuracy in defect detection, while healthcare implementations demonstrated remote patient monitoring with sub-50ms latency. Costbenefit analysis of these implementations revealed 35-45% reduced operational costs and ROI improvements up to 285% across sectors. Employing predictive modeling and trend analysis techniques, our research projects significant evolution within the next decade: 85% improvement in system autonomy, 3.5× enhancement in processing power, and 65% gains in energy efficiency. Edge computing, defined as "computation happening anywhere near the data sources and control," continues to evolve, addressing emerging challenges in distributed computing architectures.

This article presents a comprehensive approach to building enterprise-grade realtime analytics pipelines leveraging AWS managed services, specifically OpenSearch, EMR Spark, and AWS Managed Grafana. It introduces a scalable architecture that addresses the challenges of high-throughput data ingestion, low-latency processing, and interactive visualization while minimizing operational overhead. The solution

Engineering and management decision situations with more than one criterion have been encountered with increasing frequency (Adeyeye&Charles-Owaba, 2008; Adeyeye&Oyawale, 2010a; Biswas et al, 2020; Zizovic, Miljkovic, &Marinkovic, 2020). In multi-criteria decisions, cases where the criteria conflict with each other are more common. In such situations gain in one criterion leads to loss in one or more of the other criteria. Due to conflict among the criteria, it is impossible to find a point at which all the criteria would assume their optimum values simultaneously. Consequently, there is no common optimal solution. What we have is a compromise solution(s). Since the criteria remain in conflict over the decision space, the analysist often elicits and incorporates the preferences of the Decision Maker (DM) in the model such that the DM gets the best compromise solution.
The criteria are often of varying degree of importance to the DM. Hence preference elicitation is very important in multicriteria optimisation. The preferences of the DM are often expressed as weights to reflect the relative importance of the criteria. The set of weights is often referred to as preference indices or preference structure. The preference structure is very important because the compromise solution obtained often vary from one set of preference structure to another. There are three possible situations in the case of a priori articulation of DM opinion concerning the relative importance of the various criteria. The DM may be satisfied
343
with the compromise solution obtained with his preferencestructure and may not bother himself/herself about other feasible solutions that may exist. In other instance, the DM may have interest in the trade-off options available to him/her. In that case, the problem is solved repeatedly with different preference structures. In this manner, the DM learns about available trade-off options and is able to know how much he/she has to give up in one or more criterion/criteria to gain improvement in one or more of the remaining criteria. He is then able to intelligently select the most preferred solution from the candidate solutions. In some other instances, the DM may want the analyst to present the Pareto optimal solutions to him/her for evaluation after which the most preferred solution is picked. Pareto optimal solutions are often generated using the weights elicited from the DM(Adeyeye&Oyawale, 2010a, b; Adeyeye, Odu& Charles-Owaba, 2015; Zizovic, et al, 2020; Navarro, Penades-Pla, Martinez-Munoz, Rempling, &Yepes, 2020; Aiello, et al, 2020; Wang, Parhi, Rangaiah, & Jana, 2020).
Many methods have been proposed for solving multicriteria optimisation problems. Among the methods are the linear combination of objective functions or weighted-sum scalarization (WSS) method in which criteria are normalised and combined before performing optimisation of the combined objective (Adeyeye&Oyawale, 2010a;Adeyeye& Charles-Owaba, 2012; Oktal, Yaman, & Kasımbeyli, 2020 & Erozan &Calıskan, 2020). Nonpre-emptiveGoal Programming (NGP) is a distance-function approach which uses a certain target point in the decision space which represents the most desired values for the several criteria as a key element in modelling the problem (Adeyeye& Charles-Owaba, 2008; Adeyeye&Oyawale, 2010b;Bakhtavar, Prabatha, Karunathilake, Sadiq, &Hewage, 2020). Compromise Programming (CP) is another distance-function approach for which the target point is a utopian point usually not feasible which corresponds to the ideal value of each criterion (Adeyeye&Oyawale, 2010b;Adeyeye& Allu, 2017; Salman et al, 2020 and Canales,Jurasz, Beluco, &Kies, 2020). Adulbhan andTabucanon, (1977, 1979) and Chen, Wiecek, and Zhang (1998) observed that in multicriteria LP problems, the linear combination of objective functions method is very simplistic and sometimes fail to give the real compromise.
344
According to Adeyeye and Charles-Owaba (2012), this drawback could be due to the lack of sensitivity of the linear combination of objective functions.The Compromise Constraint Bicriteria LP (CCBLP) method was proposed to overcome the limitation of combination of objective functions method for bicriteria case (Adulbhan&Tabucanon, 1977 & 1979). The CCBLP hasnot gained much popularity apart from few applications probably due to lack of evidence on its efficacy (Adulbhan&Tabucanon, 1979;Adeyeye and Charles-Owaba, 2012). The DMs desire to know the relative merits of these approaches so that under any given decision situations they can be properly guided in making the choice of the method that best meet their needs. Since preference indices are often used to determine the best compromise solution, generate the Pareto optimal solutions and carry out trade-off analysis, the sensitivity of any multicriteria method to changes in weight structure could be a good approach of evaluating the usefulness of the method. In this study, the earlier work of Adeyeye and Charles-Owaba (2012) is extended by comparing the CCBLP method with three commonly used traditional weighted criteria methods, namely; WSS, NGP and CP on their sensitivities to changes in the weight structure for bicriteria LP problem.

Introduction Several tools for pain measurement including a Visual Analogue Scale (VAS) and a Verbal Rating Scale (VRS) are currently used in patients with chronic pain. The aim of the present study was to determine which of these two pain tests performs optimally in patients following groin hernia repair. Patients and methods A questionnaire identiWed pain level in a cohort of patients that had previously undergone corrective groin hernia surgery. Current pain intensity was graded on a four-point VRS scale (no pain, mild, moderate or severe pain) and on a 100-mm VAS scale (0 = no pain, 100 = unbearable). "Scale failure" (one or both tests not completed correctly) was determined, and cut-oV points for the VAS test were calculated by creating the optimum kappa coeYcient between both tools. Results The response rate was 78.2% (706/903). Scale failure was present in VAS tests more than VRS (VAS: 12.5%, 88/706 vs. VRS: 2.8%, 20/706; P < 0.001). Advanced age was a risk factor for scale failure (P < 0.001). The four categories of VRS corresponded to mean VAS scores of 1, 20, 42, and 78 mm, respectively. VAS categories associated with the highest kappa coeYcient (k = 0.78) were as follows: 0-8 = no pain, 9-32 = mild, 33-71 = moderate, >71 = severe pain. VAS scores grouped per VRS category showed considerable overlap. Age and sex did not signiWcantly inXuence cut-oV points. Conclusions Because of lower scale failure rates and overlapping VAS scores per VRS category, the VRS should be favored over the VAS in future postherniorrhaphy pain assessment. If VAS is preferred, the presented cut-oV points should be utilized. This study was unfunded.

Introduction Several tools for pain measurement including a Visual Analogue Scale (VAS) and a Verbal Rating Scale (VRS) are currently used in patients with chronic pain. The aim of the present study was to determine which of these two pain tests performs optimally in patients following groin hernia repair. Patients and methods A questionnaire identiWed pain level in a cohort of patients that had previously undergone corrective groin hernia surgery. Current pain intensity was graded on a four-point VRS scale (no pain, mild, moderate or severe pain) and on a 100-mm VAS scale (0 = no pain, 100 = unbearable). "Scale failure" (one or both tests not completed correctly) was determined, and cut-oV points for the VAS test were calculated by creating the optimum kappa coeYcient between both tools. Results The response rate was 78.2% (706/903). Scale failure was present in VAS tests more than VRS (VAS: 12.5%, 88/706 vs. VRS: 2.8%, 20/706; P < 0.001). Advanced age was a risk factor for scale failure (P < 0.001). The four categories of VRS corresponded to mean VAS scores of 1, 20, 42, and 78 mm, respectively. VAS categories associated with the highest kappa coeYcient (k = 0.78) were as follows: 0-8 = no pain, 9-32 = mild, 33-71 = moderate, >71 = severe pain. VAS scores grouped per VRS category showed considerable overlap. Age and sex did not signiWcantly inXuence cut-oV points. Conclusions Because of lower scale failure rates and overlapping VAS scores per VRS category, the VRS should be favored over the VAS in future postherniorrhaphy pain assessment. If VAS is preferred, the presented cut-oV points should be utilized. This study was unfunded.

With the introduction of electric vehicles in the automobile market, limited information is available on how the battery's energy consumption is distributed. This paper focuses on the energy consumption of the vehicle when the heating and cooling system is in operation. On average, 18 and 14% for the battery's energy capacity is allocated to heating and cooling requirements, respectively. The conventional internal combustion engine vehicle uses waste heat from its engine to provide for passenger thermal requirements at no cost to the vehicle's propulsion energy demands. However, the electric vehicle cannot avail of this luxury to recycle waste heat. In order to reduce the energy consumed by the climate control system, an analysis of the temperature profile of a vehicle's cabin space under various weather conditions is required. The present study presents a temperature predicting algorithm to predict temperature under various weather conditions. Previous studies have limited consideration to the fluctuation of solar radiation space heating to a vehicle's cabin space. This model predicts solar space heating with a mean bias error and root mean square error of 0.26 and 0.57 C, respectively. This temperature predicting model can potentially be developed with further research to predict the energy required by the vehicle's primary lithium-ion battery to heat and cool the vehicle's cabin space. Thus, this model may be used in a route planning application to reduce range anxiety when drivers undertake a journey under various ambient weather conditions while optimising the energy consumption of the electric vehicle.

The evolving digital landscape poses significant challenges to enterprises, such as managing identities and controlling access across a wide array of applications, systems, and networks. Advanced Identity Governance and Administration (IGA) platforms, such as SailPoint IdentityNow, address these challenges by delivering automated, scalable, and secure access management solutions. This paper examines SailPoint IdentityNow's architecture, features, and implementation strategies, focusing on how it strengthens identity governance, ensures compliance, and enhances operational efficiency. Real-world use cases and a roadmap for future improvements in identity governance are also discussed.

In this paper, two new architectures for high-speed CMOS wave-pipelined current-mode A/D converters (WP-IADCs) are proposed and analyzed. In the new WP-IADC architectures, the wave-pipelined theory is applied to both pipeline structures, called full WP-IADC (FWP-IADC) and indirect transfer WP-IADC (ITWP-IADC). In the FWP-IADC, each stage uses the full current-mode wave-pipelined structure without switched-current cell circuits. In the ITWP-IADC, the switched-current cells are incorporated into the wave-pipelined stages which are divided into several sections with controlled clocks. Therefore, the proposed ITWP-IADC performs optimally in terms of speed and accuracy in the WP-IADCs. Generally, the proposed WP-IADCs have the advantages of high speed, high input frequency, high efficiency of timing usage, high clock-period flexibility in switched-current cells for precision enhancement, and reduced number of switched-current cells in the overall data path for linearity improvement. According to the theoretical analysis on the proposed WP-IADC structures, the minimum sampling clock period is proportional to the intrinsic delay of the current mirror and the increased rise/fall time in each wave-pipelined stage. The HSPICE simulation results reveal that, under Nyquist rate sampling in 8-b resolution, a sampling rate of 20 and 54 MHz can be achieved for FWP-IADC and two-section ITWP-IADC, respectively. If four wave-pipelined sections are used, the ITWP-IADC can be operated at 166 MHz at an input frequency of 8 MHz. To experimentally verify the correct function of the proposed WP-IADC structures, the proposed new architecture of the FWP-IADC is implemented by using 0.35-m CMOS technology. The measurement results successfully demonstrate the feasibility of wave-pipelined IADC architectures in applications of high-speed ADCs.

In this paper, two new architectures for high-speed CMOS wave-pipelined current-mode A/D converters (WP-IADCs) are proposed and analyzed. In the new WP-IADC architectures, the wave-pipelined theory is applied to both pipeline structures, called full WP-IADC (FWP-IADC) and indirect transfer WP-IADC (ITWP-IADC). In the FWP-IADC, each stage uses the full current-mode wave-pipelined structure without switched-current cell circuits. In the ITWP-IADC, the switched-current cells are incorporated into the wave-pipelined stages which are divided into several sections with controlled clocks. Therefore, the proposed ITWP-IADC performs optimally in terms of speed and accuracy in the WP-IADCs. Generally, the proposed WP-IADCs have the advantages of high speed, high input frequency, high efficiency of timing usage, high clock-period flexibility in switched-current cells for precision enhancement, and reduced number of switched-current cells in the overall data path for linearity improvement. According to the theoretical analysis on the proposed WP-IADC structures, the minimum sampling clock period is proportional to the intrinsic delay of the current mirror and the increased rise/fall time in each wave-pipelined stage. The HSPICE simulation results reveal that, under Nyquist rate sampling in 8-b resolution, a sampling rate of 20 and 54 MHz can be achieved for FWP-IADC and two-section ITWP-IADC, respectively. If four wave-pipelined sections are used, the ITWP-IADC can be operated at 166 MHz at an input frequency of 8 MHz. To experimentally verify the correct function of the proposed WP-IADC structures, the proposed new architecture of the FWP-IADC is implemented by using 0.35-m CMOS technology. The measurement results successfully demonstrate the feasibility of wave-pipelined IADC architectures in applications of high-speed ADCs.

Stochastic optimization is used in several high impact contexts to provide optimal solutions in the face of uncertainties. This paper explores the parallelization of twostage stochastic resource allocation problems that seek an optimal solution in the first stage, while accounting for sudden changes in resource requirements by evaluating multiple possible scenarios in the second stage. Unlike typical scientific computing algorithms, linear programs (which are the individual grains of computation in our parallel design) have unpredictable and long execution times. This confounds both a priori load distribution as well as persistence-based dynamic load balancing techniques. We present a master-worker decomposition coupled with a pull-based work assignment scheme for load balance. We discuss some of the challenges encountered in optimizing both the master and the worker portions of the computations, and techniques to address them. Of note are cut retirement schemes for balancing memory requirements with duplicated worker computation, and scenario clustering for accelerating the evaluation of similar scenarios. We base our work in the context of a real application: the optimization of US military aircraft allocation to various cargo and personnel movement missions in the face of uncertain demands. We demonstrate scaling up to 122 cores of an intel R 64 cluster; even for very small, but representative datasets. Our decision to eschew problem-specific decompositions has resulted in a parallel infrastructure that should be easily adapted to other similar problems. Similarly, we believe the techniques developed in this paper will be generally applicable to other contexts that require quick solutions to stochastic optimization problems.

Stochastic optimization is used in several high impact contexts to provide optimal solutions in the face of uncertainties. This paper explores the parallelization of twostage stochastic resource allocation problems that seek an optimal solution in the first stage, while accounting for sudden changes in resource requirements by evaluating multiple possible scenarios in the second stage. Unlike typical scientific computing algorithms, linear programs (which are the individual grains of computation in our parallel design) have unpredictable and long execution times. This confounds both a priori load distribution as well as persistence-based dynamic load balancing techniques. We present a master-worker decomposition coupled with a pull-based work assignment scheme for load balance. We discuss some of the challenges encountered in optimizing both the master and the worker portions of the computations, and techniques to address them. Of note are cut retirement schemes for balancing memory requirements with duplicated worker computation, and scenario clustering for accelerating the evaluation of similar scenarios. We base our work in the context of a real application: the optimization of US military aircraft allocation to various cargo and personnel movement missions in the face of uncertain demands. We demonstrate scaling up to 122 cores of an intel R 64 cluster; even for very small, but representative datasets. Our decision to eschew problem-specific decompositions has resulted in a parallel infrastructure that should be easily adapted to other similar problems. Similarly, we believe the techniques developed in this paper will be generally applicable to other contexts that require quick solutions to stochastic optimization problems.

This comprehensive article explores the evolution and current state of performance optimization in modern technological systems, focusing on the transformative impact of generative AI and quantum computing. The article examines the progression from traditional manual optimization methods to advanced AI-driven approaches, highlighting improvements in data center operations, enterprise systems, and deep learning architectures. The article investigates the emergence of autonomous systems, adaptive learning capabilities, and the integration of quantum computing in optimization processes. Furthermore, it addresses critical ethical considerations in AI-driven optimization, including transparency, human oversight, and bias mitigation, while emphasizing the importance of balanced governance frameworks for sustainable technological advancement.

Partial anterior cruciate ligament (ACL) tears were first described nearly fifty years ago but the optimal treatment for these injuries continues to be a subject of considerable debate. A question remains whether it is advantageous to preserve the ACL remnant and augment it with a graft, or to debride it and proceed with a standard ACL reconstruction unhindered by remnant fibers in the notch. Clinical outcomes of bundle preserving surgery are promising. An increasingly large body of scientific evidence suggests that augmenting the intact bundle is beneficial in terms of vascularity, proprioception and kinematics. With this knowledge, a number of surgeons have developed techniques to augment the intact bundle of the ACL in partial tears and to biologically enhance standard reconstruction techniques by preserving the ACL remnant. Correct tunnel placement is critical for achieving successful short and long-term outcomes after ACL reconstruction. However, published studies have several limitations including a limited number of patients and lack of control groups for direct comparison of outcomes. Concerns continue to exist with respect to an increased risk of impingement following augmentation, responsible of cyclops syndrome. The objective of this article was to outline the diagnostic approach, describe a reproducible and simple surgical procedure that allows correct femoral tunnel placement without the need for aggressive notch debridement and report the clinical outcome of partial ACL reconstruction.

Implementations of the developed ML models are related to important questions of scale, resource, and management. This work investigates the use of serverless cloud models to address these issues and enhance and optimize the deployment, scalability, and maintenance of ML models. Reviewing main serverless platforms and their compatibility with different ML development and usage stages, the research compares the effectiveness, cost, and adaptability to the common application deployment practices. The study in this paper employs case and performance analysis to show and explain how serverless solutions can cut infrastructure costs and reduce the need for scaling and maintenance, among others. This paper outlines guidelines for implementing serverless technologies in ML applications and areas of concern that organizations might expect. Consequently, this research adds to the existing literature on deploying ML-based applications in the cloud while providing useful findings for developers and organizations interested in efficient, costeffective solutions.

The rapid evolution of data processing demands has led to innovative approaches in enterprise-scale data anonymization and protection. This comprehensive examination explores the implementation of Delphix across diverse cloud environments, focusing on its technical architecture, performance metrics, and compliance features. The platform demonstrates exceptional capabilities in handling sensitive data through advanced machine learning algorithms and sophisticated processing pipelines. The architecture incorporates robust security mechanisms, parallel processing capabilities, and intelligent resource optimization across multiple geographical regions. Integration with major cloud providers enables seamless scalability while maintaining strict data protection standards. The implementation showcases significant improvements in processing efficiency, reduced data breach risks, and enhanced compliance adherence through automated controls. Best practices and deployment guidelines ensure optimal performance through carefully calibrated infrastructure requirements and monitoring systems. The solution addresses the critical challenges of data privacy and security while maintaining high throughput rates and system availability across distributed environments.

Optimizing query processing in distributed database system is an important research area considering the volume of data and information being processed these days. Many techniques have been proposed for optimizing query processing in distributed databases. In this paper, we proposed a combination of two of the most commonly used techniques for optimizing query: datashipping and query-shipping techniques. This hybridized technique provides solution for storing and processing data and information for quick retrieval of information in a distributed database when the data to be retrieved are not located within a single computer system. Using this technique, each employee in an organization that is geographically located in different regions can decide to hide information about any of the employee whose data is in the organization's database by preserving their individual query intent. In the client machine, the data-shipping technique is used to help in the local processing and communication cost while the query-shipping technique is used in the server machine to optimize the given data selection. In some cases data management (update) can also be carried out by the database administrator thus the technique combines the features of both the data-shipping and query-shipping techniques. We compared the results obtained with previous results when the techniques were separately used and we discovered that our technique performs better in terms of the time and complexities of the algorithm used.

Leveraging the Azure Well-Architected Framework (WAF) for enterprise cloud governance provides organizations with a comprehensive set of best practices and tools to optimize cloud operations while ensuring security, compliance, and cost efficiency. The WAF is structured around five key pillarscost optimization, security, reliability, performance efficiency, and operational excellence-which collectively guide enterprises in building and managing cloud architectures that align with business objectives and governance requirements. This framework offers a proven methodology for managing cloud resources at scale, ensuring high availability, reducing risks, and preventing overspending. For enterprises, adopting the WAF helps streamline governance by enabling consistent policy enforcement, automated resource management, and continuous monitoring. The integration of Azure Policy, Azure Blueprints, and tools like Azure Cost Management and Azure Security Center ensures that cloud environments are governed in a way that maximizes value while adhering to regulatory standards and minimizing vulnerabilities. Furthermore, the framework supports the migration to and operation of multi-cloud and hybrid environments, providing a cohesive governance strategy across complex cloud landscapes. By implementing Azure's Well-Architected Framework, organizations can address key challenges in cloud governance-such as ensuring compliance, optimizing costs, managing performance, and securing data-while maintaining operational flexibility and agility. This research explores how leveraging the WAF can enable enterprises to create a resilient, cost-effective, and secure cloud environment, providing a foundation for both innovation and compliance in today's dynamic digital landscape.

Vertical network slicing is an innovative technology that enhances 5G networks by enabling the efficient management and optimization of data traffic across diverse industries. This paradigm divides the physical network infrastructure into multiple virtual slices, each tailored to meet specific application requirements, such as low latency for autonomous vehicles, high bandwidth for ultra-high-definition video streaming, or massive connectivity for IoT ecosystems. By leveraging advanced resource allocation algorithms, network function virtualization (NFV), and software-defined networking (SDN), vertical slicing ensures seamless adaptability to dynamic workloads and service-level agreements (SLAs).

This study explores the role of vertical network slicing in improving 5G network performance, focusing on its capability to address the challenges of traffic congestion, latency, and resource utilization. A comparative analysis of traditional static allocation methods versus slicing-based approaches highlights the latter's ability to deliver greater efficiency and service differentiation. Key findings emphasize the transformative potential of vertical network slicing in empowering telecommunications providers to support emerging use cases while maintaining operational excellence. The paper concludes with a discussion on the future scope of slicing, including integration with AI-driven orchestration and its implications for 6G readiness.

Real-time analytics has long held the promise of understanding customers, systems, and computing on  the  fly,  unlocking  new  insights  and  data-driven  decision-making.  However,  advancements  in data processing, specifically faster, cheaper, and smarter storagefor big data, have yet to be realized to  make  real-time  analytical  systems  a  reality.  This  paper  shows  how  recent  advancements  in storage technologies have begun to change the game with three critical developments. The first is the  development  of  byte-addressable  persistent  memory  that  serves  as  high-speed  storage.  The second  technology  is  direct  attached  storage  that  minimizes  the  CPU  stack  load  when  accessing data on the device, keeping more of the computer in the data store. Lastly, stateless computational storage targets distributed AI systems, pushing AI computations closer to the data source. These high-speed  storage  innovations  have  already  begun  to  be  integrated  into  cloud  frameworks, enabling storage stacks to manage these devices for users and creating a roadmap for integrating stateless  computing  storage  as  that  sector  matures.  The  paper  outlines  these  AI-centric  storage designs, explains the coming transformations to cloud computing that these storage innovations are enabling, details their status, and describes the kinds of applications that will improve with these storage designs. These changes are released from two perspectives, one for AI and data processing in the cloud, sustaining data analytics, and another for stream processing in time-series systems. Of particular importance in these designs are the storage controllers, which will be an integral part of AI-optimized data storage critical for real-time and edge settings, showing the latest storage that aims to push computation toward data and keep more of the bits intact as data are transformed at the edge with minimally powerful AI accelerators.

In presenting this thesis in partial fulfillment of the requirements for the degree of Master of Science at the University of Saskatchewan, the author agrees that the Libraries of this University may make this thesis readily available for inspection. The author further agrees that permission for copying of this thesis in any manner, in whole or in part, for scholarly purposes may be granted by the professors who supervised the thesis work or, in their absence, by the Head of the Department or the Dean of the College ofKinesiology.lt is understood that any copying or publication or use of this thesis or parts thereof for fmancial gain shall not be allowed without written permission of the author. It is also understood that due recognition shalll?e given to the author and to the University of Saskatchewan in any scholarly use which may be made of any material in this thesis.

The practical aspects of rear pick arrangements on boom-type tunnelling machine cutting heads,
investigated during long-term underground trials with the use of various types of picks and cutting heads
are described. The results indicate that with improperly positioned rear picks the cutting rate is
significantly reduced, and that progressive wear develops both on tool holders and cutting boom
components. It is also emphasised that the cutting head connection to the boom is greatly influenced by
excessive depth of sump, and the flange-type cutting heads are much more prone to wear.

An optimal performance analysis for an equivalent Carnot-like cycle heat engine of a parabolictrough direct-steam-generation solar driven Rankine cycle power plant at maximum power and maximum power density conditions is performed. Simultaneous radiation-convection and only radiation heat transfer mechanisms from solar concentrating collector, which is the high temperature thermal reservoir, are considered separately. Heat rejection to the low temperature thermal reservoir is assumed to be convection dominated. Irreversibilities are taken into account through the finite-rate heat transfer between the fixed temperature thermal reservoirs and the internally reversible heat engine. Comparisons proved that the performance of a solar driven Carnot-like heat engine at maximum power density conditions, which receives thermal energy by either radiation-convection or only radiation heat transfer mechanism and rejects its unavailable portion to surroundings by convective heat transfer through heat exchangers, has the characteristics of (1) a solar driven Carnot heat engine at maximum power conditions, having radiation heat transfer at high and convective heat transfer at low temperature heat exchangers respectively, as the allocation parameter takes small values, and of (2) a Carnot heat engine at maximum power density conditions, having convective heat transfer at both heat exchangers, as the allocation parameter takes large values. Comprehensive discussions on the effect of heat transfer mechanisms are provided.

The development of the Keck All sky Precision Adaptive optics (KAPA) project was initiated in September 2018 to upgrade the Keck I adaptive optics (AO) system to enable laser tomography adaptive optics (LTAO) with a four laser guide star (LGS) asterism. The project includes the replacement of the existing LMCT laser with a Toptica laser, the implementation of a new real-time controller (RTC) and wavefront sensor optics and camera, and a new daytime calibration and test platform to provide the required infrastructure for laser tomography. The work presented here describes the new daytime calibration infrastructure to test the performance for the KAPA tomographic algorithms. This paper outlines the hardware infrastructure for daytime calibration and performance assessment of tomographic algorithms. This includes the implementation of an asterism simulator having fiber-coupled light sources simulating four Laser Guide Stars (LGS) and two Natural Guide Stars (NGS) at the AO bench focus, as well as the upgrade of the existing TelSim on the AO bench to simulate focal anisoplanatism and wind driven atmospheric turbulence. A phase screen, that can be adjusted in effective altitude, is used to simulate wind speeds up to 10 m/s for a duration of upto 3 s.

Over the last few decades, controllers found in the industries are mostly PID controllers. They have found large recognition and applications in several industries. The majority of the controllers used in process control are of PID type. The control loops which are not properly tuned, give reduced output and inappropriate and undesired performance. There are continuous researches going on to develop new methods for PID tuning and designing. A number of algorithms have been developed by scientists and suitable methods depending upon the applications are adopted by the industries for PID tuning and designing. Several tuning methods are present but they have some restrictions. These methods do not give desired tuning parameterizations for the control systems which have higher order and delay systems. The method which is named as Dominant pole placement method provides better tuning parameterizations for the above mentioned type of systems. In this method a pair of desired poles is chos...

Efficient Dynamic Kinetic Resolution of Secondary Amines with Pd on Alkaline Earth Salts and a Lipase. -Pd immobilized on supports such as BaSO4 is an efficient heterogeneous catalyst for the racemization of chiral benzylic amines (I) and (IV). The process can be combined with enzymatic kinetic resolution in one-pot process leading to enantiomerically pure amides from the racemic amines with good yields. -(PARVULESCU, A.

Modern platform-based design involves the application-specific extension of embedded processors to fit customer requirements. To accomplish this task, the possibilities offered by recent custom/extensible processors for tuning their instruction set and microarchitecture to the applications of interest have to be exploited. A significant factor often determining the success of this process is the utomation available in application analysis and custom instruction generation. In this paper we present YARDstick, a design automation tool for custom processor development flows that focuses on generating and evaluating application-specific hardware extensions. YARDstick is a building block for ASIP development, integrating application analysis, custom instruction generation and selection with user-defined compiler intermediate representations. In a YARDstick-enabled environment, practical issues in traditional ASIP design are confronted efficiently; the exploration infrastructure is liberated from compiler and simulator idiosyncrasies, since the ASIP designer is empowered with the freedom of specifying the target architectures of choice and adding new implementations of analyses and custom instruction generation/selection methods. To illustrate the capabilities of the YARDstick approach, we present interesting exploration scenarios: quantifying the effect of machine-dependent compiler optimizations and the selection of the target architecture in terms of operation set and memory model on custom instruction generation/selection under different input/output constraints.

MPI Alltoall is one of the most communication intense collective operation used in many parallel applications. Recently, the supercomputing arena has witnessed phenomenal growth of commodity clusters built using InfiniBand and multi-core systems. In this context, it is important to optimize this operation for these emerging clusters to allow for good application scaling. However, optimizing MPI Alltoall on these emerging systems is not a trivial task. InfiniBand architecture allows for varying implementations of the network protocol stack. For example, the protocol can be totally on-loaded to a host processing core or it can be off-loaded onto the NIC or can use any combination of the two. Understanding the characteristics of these different implementations is critical in optimizing a communication intense operation such as MPI Alltoall. In this paper, we systematically study these different architectures and propose new schemes for MPI Alltoall tailored to these architectures. Specifically, we demonstrate that we cannot use one common scheme which performs optimally on each of these varying architectures. For example, on-loaded implementations can exploit multiple cores to achieve better network utilization, and in offload interfaces aggregation can be used to avoid congestion on multi-core systems. We employ shared memory aggregation techniques in these schemes and elucidate the impact of these schemes on multi-core systems. The proposed design achieves a reduction in MPI Alltoall time by 55% for 512Byte messages and speeds up the CPMD application by 33%.

This research aims at examining the strategies of performance tuning in cloud computing with emphasis on the optimization of applications, minimized response time, and optimal, affordable resource utilization. The research therefore includes a systematic literature review together with quantitative findings through empirical testing of the proposed model on Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP), in addition to qualitative insights of experts. Auto-scaling, load balancing caching, database optimizing, integration with edge computing and predictive workloads using Artificial Intelligence are the aspects that are also studied as key performance tuning latter. Soon, the investigation, which was made based on the results from applying the four techniques on different applications and two clouds, demonstrates the strength of each technique in achieving different goals. While auto-scale and load balance feature is very helpful in control of workload fluctuations, the caching and database optimization helps in the efficient retrieval of the data. Edge computing reduces latency in response to real-time applications, and the application of artificial intelligence in workload forecast smoothes resource utilization in environments with a rapidly changing workload. Accordingly, the research has shown need for careful choosing of suitable performance-oriented interventions to enhance the application’s interactions, decrease CPU utilization, and cut costs in a cloud environment. Lastly, this work offers practical knowledge about the methods of cloud performance tuning to support better application deployment in the cloud environments....

Abstract— in this paper a reliable framework for scientific workflow management system is presented. The reliability of workflow management systems could be guaranteed by supplying workflow engines with fault control components. A fault control component may receive ...

As emerging video coding standards, e.g. H.264, aim at high-quality video contents at low bit-rates, the encoding and decoding processes require much more computation than most existing standards do. This paper analyzes software implementation of a real-time H.264 decoder on general-purpose processors with media instructions. Specifically, we discuss how to optimize the speed of H.264 decoders on Intel Pentium 4 processors. This paper first analyzes the reference implementation to identify the time-consuming modules. Our study shows that a number of components, e.g., motion compensation and inverse integer transform, are the most time-consuming modules in the H.264 decoder. Second, we present a list of performance optimization methods using media instructions to improve the efficiency of these modules. After appropriate optimizations, the decoder speed improved by more than 3x-it can decode a 720x480 resolution video sequence at 48 frames per second on 2.4GHz Intel Pentium 4 processors compared to reference software's 12 frames per second. The optimization techniques demonstrated in this paper can also be applied to other video/image processing applications. Additionally, after presenting detailed application behavior on general-purpose processors, this paper discusses a few recommendations on how to design future efficient/powerful video/image applications/standards with given hardware implications.

Modern object-oriented software commonly suffers from runtime bloat that significantly affects its performance and scalability. Studies have shown that one important pattern of bloat is the work repeatedly done to compute the same data values. Very often the cost of computation is very high and it is thus beneficial to memoize the invariant data values for later use. While this is a common practice in real-world development, manually finding invariant data values is a daunting task during development and tuning. To help the developers quickly find such optimization opportunities for performance improvement, we propose a novel run-time profiling tool, called Cachetor, which uses a combination of dynamic dependence profiling and value profiling to identify and report operations that keep generating identical data values. The major challenge in the design of Cachetor is that both dependence and value profiling are extremely expensive techniques that cannot scale to large, real-world applications for which optimizations are important. To overcome this challenge, we propose a series of novel abstractions that are applied to run-time instruction instances during profiling, yielding significantly improved analysis time and scalability. We have implemented Cachetor in Jikes Research Virtual Machine and evaluated it on a set of 14 large Java applications. Our experimental results suggest that Cachetor is effective in exposing caching opportunities and substantial performance gains can be achieved by modifying a program to cache the reported data.

Abstract—This paper describes a complete demo sce-nario where a Triple Play configuration (video, audio and data) with quality of service will be shown. This scenario could be considered as the aim of the Next Generation Networks (NGN) where a unified network is used to ...

As deep learning models scale, their training cost has surged significantly. Due to both hardware advancements and limitations in current software stacks, the need for data efficiency has risen. Data efficiency refers to the effective hiding of data access latency and the avoidance of unnecessary data movements. Major challenges arise from the growing disparity between GPU memory bandwidth and computational throughput, imminent GPU memory capacity limitations, and inefficiencies in the PyTorch software stack, including a lack of device-specific PCIe transfer optimizations and high-level domain-specific abstractions.

To effectively mitigate these data inefficiencies for deep learning training, this dissertation analyzes data inefficiency in representative deep training tasks, specifically in graph neural networks (GNNs) and large language models (LLMs). It then proposes novel runtime and code generation techniques to mitigate these challenges and implements these optimizations seamlessly within the PyTorch stack while maintaining strong programmability and interoperability.

First, PyTorch-Direct is devised to incorporate the GPU-centric PCIe data transfer paradigm in PyTorch for GNN training. PyTorch-Direct significantly reduces CPU utilization, resulting in higher end-to-end training performance. For the input datasets and GNN architectures evaluated, PyTorch-Direct decreases the overall training time by up to 38.2%.

Next, Hector intermediate representation (IR) and its code generator are proposed to introduce domain-specific high-level abstraction and systematically address memory-intensive performance challenges for relational graph neural networks (RGNNs). The performance challenges stem from RGNN's inherent memory intensiveness, the gap between the programming interface and the kernel APIs, and the high kernel optimization cost due to the kernels' coupling with layout and heterogeneity. Using a general matrix multiply (GEMM) template and a traversal template, Hector achieves up to a 43.7× speed-up in training and inference compared to the state-of-the-art systems. Linear operator reordering and compact tensor materialization further achieve up to 3.8× speed-up compared to the Hector unoptimized code.

Finally, in LLM training, the throughput has been increasingly constrained by GPU memory capacity. To mitigate this, the SSDTrain offloading framework is designed and implemented. Since activations take most of the GPU memory, SSDTrain offloads activations to Non-Volatile Memory Express (NVMe) SSDs with a direct GPU–SSD data path and good interoperability. The evaluation shows that SSDTrain reduces activations peak memory use by up to 47% with negligible overhead. We further analyze how the reduced activation memory use may be leveraged to increase throughput by increasing micro-batch size and reducing pipeline parallelism bubbles.

Together, these contributions demonstrate that code generation and runtime techniques can systematically mitigate the data management bottlenecks in deep learning training, which stem from the data-intensive nature of workloads and the oversimplification inherent in the deep learning training software stack.

In recent years, the heterogeneity of both commodity and supercomputers hardware has increased sharply. Accelerators, such as GPUs or Intel Xeon Phi co-processors, are often key to improving speed and energy efficiency of highly-parallel codes. However, due to the complexity of heterogeneous architectures, optimization of codes for a certain type of architecture as well as porting codes across different architectures, while maintaining a comparable level of performance, can be extremely challenging. Addressing the challenges associated with performance optimization and performance portability, autotuning has gained a lot of interest. Autotuning of performance-relevant source-code parameters allows to automatically tune applications without hard coding optimizations and thus helps with keeping the performance portable. In this paper, we introduce a benchmark set of ten autotunable kernels for important computational problems implemented in OpenCL or CUDA. Using our Kernel Tuning Toolkit, we show that with autotuning most of the kernels reach near-peak performance on various GPUs and outperform baseline implementations on CPUs and Xeon Phis. Our evaluation also demonstrates that autotuning is key to performance portability. In addition to offline tuning, we also introduce dynamic autotuning of code optimization parameters during application runtime. With dynamic tuning, the Kernel Tuning Toolkit enables applications to re-tune performance-critical kernels at runtime whenever needed, for example, when input data changes. Although it is generally believed that autotuning spaces tend to be too large to be searched during application runtime, we show that it is not necessarily the case when tuning spaces are designed rationally. Many of our kernels reach near peak-performance with moderately sized tuning spaces that can be searched at runtime with acceptable overhead. Finally we demonstrate, how dynamic performance tuning can be integrated into a real-world application from cryo-electron microscopy domain.

Questo manuale offre una guida su Kubernetes, esplorandone i concetti fondamentali, l'architettura distribuita e le best practice per la gestione di cluster containerizzati. Partendo da un'introduzione teorica, il testo accompagna il lettore attraverso ogni fase, dalla configurazione iniziale alla gestione avanzata delle risorse, con particolare attenzione alla sicurezza e alla scalabilità. Una sezione dedicata fornisce istruzioni pratiche per l'implementazione di Kubernetes su sistemi Linux, rendendo il manuale un riferimento per sviluppatori, amministratori di sistema e team DevOps. Arricchito da casi di studio reali e strumenti utili, è una guida completa per comprendere e implementare l'orchestrazione dei container in ambienti cloud-native.

Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g. , payment of royalties). Copies may be requested from IBM T.

In today's data-driven environment, efficient data operations are essential for organizations to optimize performance, enhance data accuracy, and enable rapid decision-making. This paper presents an innovative approach to implementing an automated data ingestion and processing framework designed to streamline repetitive tasks, ensure data quality, and support scalability within complex data ecosystems. The approach centers on a multi-step process that integrates robotic process automation (RPA), serverless computing, and advanced data transformation algorithms, thereby reducing manual interventions and accelerating data integration from multiple sources. The data ingestion process initiates with the identification and automation of repetitive data collection tasks through RPA, effectively reducing the time and potential human error associated with manual operations. Subsequently, serverless computing and platforms such as Alteryx are utilized to integrate data from diverse sources into a unified true-source repository, following either ETL (Extract, Transform, Load) or ELT (Extract, Load, Transform) workflows. This integration facilitates seamless data transformation and mapping, applying business logic and best practices to ensure alignment with organizational data standards. Automated quality monitoring is established post-ingestion to maintain high data quality, deploying event-driven triggers to detect anomalies, validate data integrity, and promptly notify relevant stakeholders of any irregularities. The technology stack supporting this framework includes Snowflake, AWS Redshift, and Azure Data Storage, along with relational databases like SQL Server and MySQL. These tools are selected for their robust processing capabilities and scalability, addressing challenges such as real-time data processing and storage requirements. Additionally, thorough documentation and version control are maintained to capture process updates and ensure a reliable knowledge base for future iterations.

Self-adaptive software systems use observations of their own behavior, and that of their environment, to select and enact adaptations in accordance with some objective(s). This adaptation is a higher-level system function that performs optimizations, manages faults, or otherwise supports achieving an objective via changes in the running system. In this paper, we show how this capability can be realized using techniques found in hierarchical control systems, and we discuss interrelated issues of stability, assurance, and implementation.