Major Cloud Service Providers Research Articles

Economics and Equity of Large Language Models: Health Care Perspective.

Large language models (LLMs) continue to exhibit noteworthy capabilities across a spectrum of areas, including emerging proficiencies across the health care continuum. Successful LLM implementation and adoption depend on digital readiness, modern infrastructure, a trained workforce, privacy, and an ethical regulatory landscape. These factors can vary significantly across health care ecosystems, dictating the choice of a particular LLM implementation pathway. This perspective discusses 3 LLM implementation pathways-training from scratch pathway (TSP), fine-tuned pathway (FTP), and out-of-the-box pathway (OBP)-as potential onboarding points for health systems while facilitating equitable adoption. The choice of a particular pathway is governed by needs as well as affordability. Therefore, the risks, benefits, and economics of these pathways across 4 major cloud service providers (Amazon, Microsoft, Google, and Oracle) are presented. While cost comparisons, such as on-demand and spot pricing across the cloud service providers for the 3 pathways, are presented for completeness, the usefulness of managed services and cloud enterprise tools is elucidated. Managed services can complement the traditional workforce and expertise, while enterprise tools, such as federated learning, can overcome sample size challenges when implementing LLMs using health care data. Of the 3 pathways, TSP is expected to be the most resource-intensive regarding infrastructure and workforce while providing maximum customization, enhanced transparency, and performance. Because TSP trains the LLM using enterprise health care data, it is expected to harness the digital signatures of the population served by the health care system with the potential to impact outcomes. The use of pretrained models in FTP is a limitation. It may impact its performance because the training data used in the pretrained model may have hidden bias and may not necessarily be health care-related. However, FTP provides a balance between customization, cost, and performance. While OBP can be rapidly deployed, it provides minimal customization and transparency without guaranteeing long-term availability. OBP may also present challenges in interfacing seamlessly with downstream applications in health care settings with variations in pricing and use over time. Lack of customization in OBP can significantly limit its ability to impact outcomes. Finally, potential applications of LLMs in health care, including conversational artificial intelligence, chatbots, summarization, and machine translation, are highlighted. While the 3 implementation pathways discussed in this perspective have the potential to facilitate equitable adoption and democratization of LLMs, transitions between them may be necessary as the needs of health systems evolve. Understanding the economics and trade-offs of these onboarding pathways can guide their strategic adoption and demonstrate value while impacting health care outcomes favorably.

Enabling high-performance cloud computing for the Community Multiscale Air Quality Model (CMAQ) version 5.3.3: performance evaluation and benefits for the user community.

The Community Multiscale Air Quality Model (CMAQ) is a local- to hemispheric-scale numerical air quality modeling system developed by the U.S. Environmental Protection Agency (USEPA) and supported by the Community Modeling and Analysis System (CMAS) center. CMAQ is used for regulatory purposes by the USEPA program offices and state and local air agencies and is also widely used by the broader global research community to simulate and understand complex air quality processes and for computational environmental fate and transport and climate and health impact studies. Leveraging state-of-the-science cloud computing resources for high-performance computing (HPC) applications, CMAQ is now available as a fully tested, publicly available technology stack (HPC cluster and software stack) for two major cloud service providers (CSPs). Specifically, CMAQ configurations and supporting materials have been developed for use on their HPC clusters, including extensive online documentation, tutorials and guidelines to scale and optimize air quality simulations using their services. These resources allow modelers to rapidly bring together CMAQ, cloud-hosted datasets, and visualization and evaluation tools on ephemeral clusters that can be deployed quickly and reliably worldwide. Described here are considerations in CMAQ version 5.3.3 cloud use and the supported resources for each CSP, presented through a benchmark application suite that was developed as an example of a typical simulation for testing and verifying components of the modeling system. The outcomes of this effort are to provide findings from performing CMAQ simulations on the cloud using popular vendor-provided resources, to enable the user community to adapt this for their own needs, and to identify specific areas of potential optimization with respect to storage and compute architectures.

Cross-disciplinary AI supply chain risk assessment

While AI remains chip based and part of both commercial and national strategic superiority goals, it is useful to examine the security and risks associated with achieving those goals. The future strategy rests perilously on an unstable inverted triangle of financial and economic reality. This paper presents the AI chip supply chain as an inverted triangle which base/apex is dependent on a global single supplier with the capability of producing equipment essential for their manufacture. It highlights the dependence on a single company for the fabrication of those chips, and the security risks associated with that supplier being Taiwanese in limited foreign ownership. It is suggested that the increasing tensions between China and the USA have resulted, in part, from this dependence, which was demonstrated by the supply chain crisis resulting from Covid-19. The attempt to reduce this dependence led to the CHIPS and Science Act 2022, signed into law by President Biden. In part of the inverted triangle are found Big Tech and the major Cloud Service Providers. They vary between 60% - 80% of their market capital being in financial institutional ownership, most of which is held by a very limited number of institutions, not all of whom are publicly quoted. To doubt the influence wielded by those financial institutions, just a single institution with major Big Tech and Cloud holdings has, at 31 December 2022, USD 8.59 trillion of assets under management. This represents economic power and places it between the equivalent Gross Domestic Product of China (USD 19.37 trillion) and Japan (USD 4.41 trillion) the second and third entries behind the USA in the GDP rankings. Financial institutions are market driven to achieve growth, contribute to economic stability, and are to an extent regulated by unelected vested interests and organisations. The battlefield for national supremacy of AI may concern chips, until the arrival of quantum AI. Current Chinese economic woes are providing the momentum for pre-emptive strikes at the semiconductor industry, and an inverted triangle is neither a secure nor stable structure for a supply chain.

Automatically Expanding User-Management System for Massive Users in the Cloud Platform

Cloud computing has become one of the key technologies used for big data processing and analytics. User management on cloud platforms is a growing challenge as the number of users and the complexity of systems increase. In light of the user-management system provided by major cloud service providers, which cannot manage multiple types of user systems, this article proposed scale-out automated expansion user management for authorization synchronization to improve the efficiency and scalability of user management on cloud platforms. Three modules for user-automated expansion were designed and implemented to synchronize the authentication information from the cloud platform resource user to the data-processing user. Additionally, an optimized dynamically weighted load-balancing algorithm in Nginx is presented in this article that adjusts the weight according to load information such as CPU and memory usage, and a better load balance can be achieved. The effectiveness of the proposed user-management system is substantiated by comparing it with two existing infrastructures, including multiple data centers and the Huawei cloud platform. The experimental results validate the finding that scale-out automated expansion user management across the Huawei cloud platform can effectively synchronize data accessing authority with cloud resource utilizing authority. Furthermore, the optimized weighted load-balancing algorithm is also valuable for massive concurrent user registration based on limited cloud resources. In the future, this scale-out user-management system could be applied to other cloud platforms and extended by database synchronization to satisfy the needs of data sharing among multiple types of users belonging to different cloud platforms.

Enhancing Data Security in Cloud Computing Using Blockchain

As cloud computing continues to proliferate, ensuring data security within cloud environments becomes increasingly critical. This paper explores the integration of blockchain technology as a means to enhance data security in cloud computing. Blockchain, with its decentralized and immutable ledger system, offers unique capabilities that address key security challenges such as data privacy, integrity, identity management, and compliance. Through a comprehensive analysis of blockchain's potential applications and integration strategies, this paper provides insights into how blockchain can bolster data security in cloud computing environments. Case studies and real-world implementations highlight successful integration efforts by major cloud service providers such as IBM, Microsoft, and Amazon. Additionally, this paper discusses the challenges and limitations associated with blockchain integration and explores future directions and emerging trends in the intersection of blockchain and cloud security. By leveraging blockchain technology, organizations can enhance their security posture in the cloud, ensuring the confidentiality, integrity, and availability of their sensitive data.

AMAZON S3 STATIC WEBSITE HOSTING

Nowadays, especially in period of energetic crisis, it is very important to get reliable data storage and uninterrupted access to websites. Considering that the world’s mainstream is using the clouds services for educational, scientific, business purposes it is important to explore features of hosting websites in major cloud service providers. One of the leader’s vendors in cloud servicing is Amazon with its Web Services (AWS). AWS offers hosting solutions such as Lightsail, Amplify Console, Simple Storage Service (S3) and Elastic Cloud Computing, from simple static websites to complicated dynamic, The article deals with the method of hosting static websites. Simple Storage Service was used for this purpose. The method has many benefits: the service is simple, cheap, scales well, manages everything and has great integration with other AWS services. This method is the best for websites that do not contain server-side scripting, like PHP or ASP.NET. It is ideal for websites that change infrequently, such as personal, promo, startup websites, websites of small businesses and organizations. The article focuses on how set permissions, properties and policies of AWS S3 bucket for hosting a website. But usage of S3 bucket alone provides only http connection. That’s why to deploy secure connection and speed up delivery of the static content was proposed to use AWS CloudFront service. CloudFront is content delivery network service, that uses the latest version of Transport Layer Security Protocol and saves files in the cache for 24 hours. The article shows how to configure CloudFront distribution to serve https requests for an Amazon S3 bucket. Configuration S3 bucket REST API endpoint was used as the most secure case. As a result, latency was reduced, security was improved due to traffic encryption and the content of S3 bucket was kept private.

A Framework for Attribute-Based Access Control in Processing Big Data with Multiple Sensitivities

There is an increasing demand for processing large volumes of unstructured data for a wide variety of applications. However, protection measures for these big data sets are still in their infancy, which could lead to significant security and privacy issues. Attribute-based access control (ABAC) provides a dynamic and flexible solution that is effective for mediating access. We analyzed and implemented a prototype application of ABAC to large dataset processing in Amazon Web Services, using open-source versions of Apache Hadoop, Ranger, and Atlas. The Hadoop ecosystem is one of the most popular frameworks for large dataset processing and storage and is adopted by major cloud service providers. We conducted a rigorous analysis of cybersecurity in implementing ABAC policies in Hadoop, including developing a synthetic dataset of information at multiple sensitivity levels that realistically represents healthcare and connected social media data. We then developed Apache Spark programs that extract, connect, and transform data in a manner representative of a realistic use case. Our result is a framework for securing big data. Applying this framework ensures that serious cybersecurity concerns are addressed. We provide details of our analysis and experimentation code in a GitHub repository for further research by the community.

The Economic Impact of Cloud Technologies on the Industry 4.0 Development

This article explores the profound economic impact of cloud technologies on Industry 4.0, focusing on how these technologies are revolutionizing the industrial sector. It delves into the transformation of IT investments from capital expenditure (CAPEX) to operational expenditure (OPEX) due to cloud computing. This shift is making advanced digital technologies more accessible and affordable, particularly for small and medium-sized enterprises (SMEs). By reducing the need for significant upfront investments and providing scalable, pay-as-you-go solutions, cloud computing significantly enhances the return on investment (ROI) for Industry 4.0 initiatives. A comparative analysis of major cloud service providers—Amazon Web Services (AWS), Google Cloud Platform (GCP), and Microsoft Azure—reveals the diverse range of services and economic benefits they offer for Industry 4.0 applications. AWS leads with comprehensive IoT and machine learning services, GCP excels in data analytics and AI capabilities, and Microsoft Azure provides robust enterprise integrations and hybrid cloud solutions. The article also addresses critical challenges such as security, compliance, and the risks associated with cloud provider lock-in. It offers strategic insights into cost management practices that can maximize the economic benefits of cloud adoption, such as leveraging multi-cloud strategies and utilizing auto-scaling and reserved instances. Furthermore, the article includes case studies from leading industrial companies like Siemens, General Electric, BMW, and ABB. These examples illustrate how cloud-based Industry 4.0 solutions enhance operational efficiency, reduce costs, and drive innovation. For instance, Siemens leverages AWS for scalable IoT solutions, GE utilizes Azure for data-driven industrial insights, and BMW employs GCP for advanced data analytics to improve manufacturing processes. In conclusion, cloud technologies are essential enablers of Industry 4.0, offering significant economic advantages and fostering innovation and efficiency. By overcoming historical barriers to entry, especially for SMEs, and providing flexible, scalable solutions, cloud computing is transforming the industrial landscape, driving growth, and facilitating the widespread adoption of advanced manufacturing technologies.

Model Protection: Real-Time Privacy-Preserving Inference Service for Model Privacy at the Edge

Major cloud service providers with well-equipped infrastructure, experienced machine learning (ML) expertise, and enriched training datasets are building ML-as-a-Service (MLaaS) systems, in which clients can query ML-based prediction services with their data. Instead of moving private data to the cloud, in this work, we design, implement, and evaluate a novel secure ML system to enable MLaaS on edge devices. To protect the proprietary ML models on edge devices from revealing to the clients while maintaining a real-time inference is challenging. Existing privacy-preserving ML techniques can hardly satisfy real-time requirements. In our solution, we employ a secure enclave (e.g., SGX) to offer security and provide better efficiency than cryptographic techniques. However, the enclave alone cannot achieve real-time capability due to its limited capacity. We observe that the ML model imposes a severe accuracy degradation when adding noise to a few model weights. Based on this, we design a suite of novel solutions to optimize the performance of secure enclave-based inference service at the edge by enclosing only <inline-formula><tex-math notation="LaTeX">$1\%$</tex-math></inline-formula> computation within secure enclaves. Our work can achieve up to a <inline-formula><tex-math notation="LaTeX">$7.8\times$</tex-math></inline-formula> increase in efficiency and a <inline-formula><tex-math notation="LaTeX">$27\times$</tex-math></inline-formula> reduction in memory usage compared to the state-of-the-art.

Demystifying the Soft and Hardened Memory Systems of Modern FPGAs for Software Programmers through Microbenchmarking

Both modern datacenter and embedded Field Programmable Gate Arrays (FPGAs) provide great opportunities for high-performance and high-energy-efficiency computing. With the growing public availability of FPGAs from major cloud service providers such as AWS, Alibaba, and Nimbix, as well as uniform hardware accelerator development tools (such as Xilinx Vitis and Intel oneAPI) for software programmers, hardware and software developers can now easily access FPGA platforms. However, it is nontrivial to develop efficient FPGA accelerators, especially for software programmers who use high-level synthesis (HLS).The major goal of this article is to figure out how to efficiently access the memory system of modern datacenter and embedded FPGAs in HLS-based accelerator designs. This is especially important for memory-bound applications; for example, a naive accelerator design only utilizes less than 5% of the available off-chip memory bandwidth. To achieve our goal, we first identify a comprehensive set of factors that affect the memory bandwidth, including (1) the clock frequency of the accelerator design, (2) the number of concurrent memory access ports, (3) the data width of each port, (4) the maximum burst access length for each port, and (5) the size of consecutive data accesses. Then, we carefully design a set of HLS-based microbenchmarks to quantitatively evaluate the performance of the memory systems of datacenter FPGAs (Xilinx Alveo U200 and U280) and embedded FPGA (Xilinx ZCU104) when changing those affecting factors, and we provide insights into efficient memory access in HLS-based accelerator designs. Comparing between the typically used soft and hardened memory systems, respectively, found on datacenter and embedded FPGAs, we further summarize their unique features and discuss the effective approaches to leverage these systems. To demonstrate the usefulness of our insights, we also conduct two case studies to accelerate the widely used K-nearest neighbors (KNN) and sparse matrix-vector multiplication (SpMV) algorithms on datacenter FPGAs with a soft (and thus more flexible) memory system. Compared to the baseline designs, optimized designs leveraging our insights achieve about\( 3.5\times \)and\( 8.5\times \)speedups for the KNN and SpMV accelerators. Our final optimized KNN and SpMV designs on a Xilinx Alveo U200 FPGA fully utilize its off-chip memory bandwidth, and achieve about\( 5.6\times \)and\( 3.4\times \)speedups over the 24-core CPU implementations.

Enhancing Security-Problem-Based Deep Learning in Mobile Edge Computing

The implementation of a variety of complex and energy-intensive mobile applications by resource-limited mobile devices (MDs) is a huge challenge. Fortunately, mobile edge computing (MEC) as a new computing paragon can offer rich resources to perform all or part of the MD’s task, which greatly reduces the energy consumption of the MD and improves the quality of service (QoS) for applications. However, offloading tasks to the edge server is vulnerable to attacks such as tampering and snooping, resulting in a deep learning (DL) security feature developed by major cloud service providers. An effective security strategy method to minimize ongoing attacks in the MEC setting is proposed. The algorithm is based on the synthetic principle of a special set of strategies, and it can quickly construct suboptimal solutions even if the number of targets achieves hundreds of millions. In addition, for a given structure and a given number of patrollers, the upper bound of the protection level can be obtained, and the lower bound required for a given protection level can also be inferred. These bounds apply to universal strategies. By comparing with the previous three basic experiments, it can be proved that our algorithm is better than the previous ones in terms of security and running time.

Attribute-Based Access Control for AWS Internet of Things and Secure Industries of the Future

Internet of Things (IoT) is revolutionizing and enhancing the quality of human lives in every aspect. With a disruption of IoT devices and applications, attackers are leveraging weak authentication and access control mechanisms on these IoT devices and applications to gain unauthorized access on user devices and data and cause them harm. Access control is a critical security mechanism to secure the IoT ecosystem which comprises cloud computing and edge computing services along with smart devices. Today major cloud and IoT service providers including Amazon Web Services (AWS), Google Cloud Platform (GCP), and Azure utilize some customized forms of Role-Based Access Control (RBAC) model along with specific authorization policies enabled by policy-based access control models. To enable fine-grained access control and overcome limitations of existing access control models, there is an imminent need to develop a flexible and dynamic access control model for securing smart devices, data and resources in the cloud-enabled IoT architecture. In this paper, we develop a formal attribute-based access control (ABAC) model for AWS IoT by building upon and extending previously developed access control model for AWS IoT, known as AWS-IoTAC model. We demonstrate the applicability of our proposed model through an industrial IoT use case and its implementation in the AWS IoT platform. Our proposed fine grained model for AWS IoT incorporates its existing capabilities and introduces new attributes for IoT entities and attribute-based policies for enabling expressive access control in AWS IoT. We also evaluate the performance of our model on the AWS cloud and IoT platform with the future smart industries use-case to depict the feasibility of our model in a real-world platform.

FPGAaaS: A Survey of Infrastructures and Systems

The popularity of cloud computing services for delivering and accessing infrastructure on demand has significantly increased over the last few years. Concurrently, the usage of FPGAs to accelerate compute-intensive applications has become more widespread in different computational domains due to their ability to achieve high throughput and predictable latency while providing programmability and improved energy efficiency. Computationally intensive applications such as big data analytics, machine learning, and video processing have been accelerated by FPGAs. With the exponential workload increase in data centers, major cloud service providers have made FPGAs and their capabilities available as cloud services. However, enabling FPGAs in the cloud is not a trivial task due to incompatibilities with existing cloud infrastructure and operational challenges related to abstraction, virtualization, partitioning, and security. In this article, we survey recent frameworks for offering FPGA hardware acceleration as a cloud service, classify them based on their virtualization mode, tenancy model, communication interface, software stack, and hardware infrastructure. We further highlight current FPGAaaS trends and identify FPGA resource sharing, security, and microservicing as important areas for future research.

A review on serverless architectures - function as a service (FaaS) in cloud computing

Emergence of cloud computing as the inevitable IT computing paradigm, the perception of the compute reference model and building of services has evolved into new dimensions. Serverless computing is an execution model in which the cloud service provider dynamically manages the allocation of compute resources of the server. The consumer is billed for the actual volume of resources consumed by them, instead paying for the pre-purchased units of compute capacity. This model evolved as a way to achieve optimum cost, minimum configuration overheads, and increases the application's ability to scale in the cloud. The prospective of the serverless compute model is well conceived by the major cloud service providers and reflected in the adoption of serverless computing paradigm. This review paper presents a comprehensive study on serverless computing architecture and also extends an experimentation of the working principle of serverless computing reference model adapted by AWS Lambda. The various research avenues in serverless computing are identified and presented.

Bulk Savings for Bulk Transfers: Minimizing the Energy-Cost for Geo-Distributed Data Centers

With the fast proliferation of cloud computing, major cloud service providers, e.g., Amazon, Google, Facebook, etc., have been deploying more and more geographically distributed data centers to provide customers with better reliability and quality of services. A basic demand in such a geo-distributed data center system is to transfer bulk volumes of data from one data center to another. Geographic distribution and large delay-tolerance of such inter-data-center bulk data transfers provide cloud service providers opportunities to optimize the operating cost. Most existing studies on inter-data-center bulk data transfers focus on minimizing the network bandwidth cost. However, the energy-cost of the bulk data transfers, which also accounts for a large proportion of operating cost in the data centers, still remains unexplored. This is an important problem, especially in the multi-electricity-market environment, where the electricity price exhibits both spatial and temporal diversities. In this paper, we systematically study the problem of how to route and schedule inter-data-center bulk data transfers to minimize the energy-cost for geo-distributed data centers. We model this problem as a min-cost multi-commodity flow problem and develop an efficient two-stage optimization method to solve it. Extensive evaluations with real-life inter-data-center network and electricity prices show that our method brings significant energy-cost savings over existing bulk data transfer methods.

Continuous memory allocation model for cloud services

Cloud computing has been growing at a staggering rate by offering a flexible and financially attractive venue for businesses and consumers. Within this context, memory allocation has a significant bearing on cloud-based services. Currently, all major cloud service providers support a small set of discrete memory sizes. We propose a cloud computing service that advertises continuous (any) memory request size, while actually supporting a small number of quantized memory sizes. This scheme redefines and transforms the manner in which cloud services are offered to the public by simplifying the ever-increasing level of pricing complexity. Our proposal targets the root causes of complexity. A service provider with a continuous service model will have a distinct advantage over the competition. We utilize mathematical algorithms to quantize and map the continuous (any) memory request sizes into a small number of quantized sizes optimally, with minimal loss. Furthermore, we investigate different factors affecting the continuous model, such as worth structure, size request distribution, total memory size, and granularity. A simulation is used to conduct our study and confirm our findings.

Energy efficient computing, clusters, grids and clouds: A taxonomy and survey

Cloud computing continues to play a major role in transforming the IT industry by facilitating elastic on-demand provisioning of computational resources including processors, storage and networks. This is necessarily accompanied by the creation, and refreshes, of large-scale systems including cluster, grids and datacenters from which such resources are provided. These systems consume substantial amounts of energy, with associated costs, leading to significant CO2 emissions. In 2014, these systems consumed 70billion kWh of energy in US; this is 1.8% of the US total energy consumption, and future consumption is expected to continue around this level with approximately 73billion kWh by 2020. The energy bills for major cloud service providers are typically the second largest item in their budgets due to the increased number of computational resources. Energy efficiency in these systems serves the providers interests in saving money to enable reinvestment, reduce supply costs and also reduces CO2 emissions. In this paper, we discuss energy consumption in large scale computing systems, such as scientific high performance computing systems, clusters, grids and clouds, and whether it is possible to decrease energy consumption without detrimental impact on service quality and performance. We discuss a number of approaches, reported in the literature, that claim to improve the energy efficiency of such large scale computing systems, and identify a number of open challenges. Key findings include: (i) in clusters and grids, use of system level efficiency techniques might increase their energy consumption; (ii) in (virtualized) clouds, efficient scheduling and resource allocation can lead to substantially greater economies than consolidation through migration; and (iii) in clusters, switching off idle resources is more energy efficient, however in (production) clouds, performance is affected due to demand fluctuation.

Topological analysis of cloud service connectivity

Relying on services in the cloud involves manifold availability risks and concerns. This article focuses on the network reachability of cloud services. We present a study on cloud outages and causes, and analyze the topological connectivity of major cloud service providers (CSPs) by graph-based measures. Our approach is based on the construction and integration of an empirical dataset describing the connections between Autonomous Systems (ASs) of organizations that form the Internet backbone. According to our findings, though the ASs of CSPs generally appear to be better connected than an average AS, they also vastly differ in several connectivity measures, sometimes by more than an order of magnitude. Our results help to identify well-connected CSPs and CSPs that could potentially suffer more from Internet outages, if no additional path redundancy is provided. Our approach can be used by CSPs to assess connectivity beyond their own premises. It can also support cloud service customers during benchmarking and selection of CSPs when high availability is a critical requirement.

An Optimization Algorithm for Selecting Cloud Service Partner Based on Graph Theory

The cloud computing market is composed by many cloud service providers, It is a key problem about how to choose and to determine the appropriate the competitiveness partners between the major cloud service providers. For this problem we propose a optimization algorithm based on graph theory solution. First we should clearly consider the factors in choosing the partner's .On this basis we establish the mathematical model of the cloud service provider partners. How to select the tender of cloud service providers in the problem, the paper gives the optimization algorithm based on graph theory, last we give an example to verify the effectiveness of the algorithm.