Source-linked AI summary
Modern Computing: Vision and Challenges
Sukhpal Singh Gill, Huaming Wu, Panos Patros, Carlo Ottaviani, Priyansh Arora, Victor Casamayor Pujol, David Haunschild, Ajith Kumar Parlikad, Oktay Cetinkaya, Hanan Lutfiyya, Vlado Stankovski, Ruidong Li, Yuemin Ding, Junaid Qadir, Ajith Abraham, Soumya K. Ghosh, Houbing Herbert Song, Rizos Sakellariou, Omer Rana, Joel J. P. C. Rodrigues, Salil S. Kanhere, Schahram Dustdar, Steve Uhlig, Kotagiri Ramamohanarao, Rajkumar Buyya
TL;DR
Modern computing systems face challenges from distributed data, large-scale resource management, and expanding infrastructure demands. This review traces computing's evolution across architectures and paradigms, identifying specialized systems and systemic resource-management approaches as central developments.
Problem
Modern distributed computing systems must manage data across multiple infrastructures while meeting latency, dependability, cost-effectiveness, throughput, and sustainability requirements.
Method
The paper comprehensively reviews computing-system evolution, architectures, paradigms, and resource-management approaches across centralized, distributed, cloud, fog, and related systems.
Results
The review identifies a trajectory toward specialized systems and models, decentralized structures, dynamic resource adaptation, and systemic approaches to resource management.
Takeaways & Limitations
Modern computing research must consider software, networking, server architecture, ventilation, and AI-assisted resource management together.
Takeaways & Limitations
Blockchain-based fog architectures still require efficient consensus techniques that reduce block sharing and copying.
Abstract
from arXiv · showhide
Over the past six decades, the computing systems field has experienced significant transformations, profoundly impacting society with transformational developments, such as the Internet and the commodification of computing. Underpinned by technological advancements, computer systems, far from being static, have been continuously evolving and adapting to cover multifaceted societal niches. This has led to new paradigms such as cloud, fog, edge computing, and the Internet of Things (IoT), which offer fresh economic and creative opportunities. Nevertheless, this rapid change poses complex research challenges, especially in maximizing potential and enhancing functionality. As such, to maintain an economical level of performance that meets ever-tighter requirements, one must understand the drivers of new model emergence and expansion, and how contemporary challenges differ from past ones. To that end, this article investigates and assesses the factors influencing the evolution of computing systems, covering established systems and architectures as well as newer developments, such as serverless computing, quantum computing, and on-device AI on edge devices. Trends emerge when one traces technological trajectory, which includes the rapid obsolescence of frameworks due to business and technical constraints, a move towards specialized systems and models, and varying approaches to centralized and decentralized control. This comprehensive review of modern computing systems looks ahead to the future of research in the field, highlighting key challenges and emerging trends, and underscoring their importance in cost-effectively driving technological progress.
1. Introduction
This article reviews modern computing by tracing paradigm shifts, technology drivers, and changing system structures. It contributes a taxonomy, performance criteria, open challenges, and emerging-trend analysis.
- The study suggests that next-generation systems will decentralize computational services through workload-specific decentralized tools targeting capacity and power accessibility.
- The study traces computing evolution from early to modern systems, focusing on technological drivers from 1960–2023.
- It develops a taxonomy based on computing focus or paradigms, technologies or impact areas, and trends or observations.
- The classification covers standalone versus networked, general-purpose versus specialized, centralized versus decentralized, emerging technologies, and parallel versus sequential computing.
- The article identifies impact and performance criteria spanning performance, efficiency, social impact, security and compliance, and economics and management.
- It identifies open challenges and research directions, summarizes computing traits and resources, and introduces a hype cycle for emerging trends.
2. Early Computing to Modern Computing: A Vision
Computing systems have progressed from early architectures toward contemporary systems as technological advances optimize available hardware and replace earlier state-of-the-art models. Figure 1 presents this transition as a progression of modern computing technology.
- Over six decades, novel computing models and innovations have successively replaced previous state-of-the-art systems while optimizing available hardware.
- Early computing systems were limited to one task per system, requiring multiple systems to work together for broader tasks.
3. Evolution of Computing Paradigms: Technological Drivers
Computing paradigms evolved from centralized and specialized systems toward models that distribute workloads, pool resources, and support autonomous or mobile use. These shifts were driven by changing performance, connectivity, and application requirements.
- Centralized to distributed computing: Computing progressed from centralized client-server platforms to increasingly distributed paradigms that share workloads across providers, consumers, nodes, or peers.The client-server model used servers, networking, and load balancing, while later models distributed computation across clusters, grids, and peer networks.
- High-capacity systems: Supercomputers and mainframes addressed computationally intensive and data-intensive workloads, while virtualization improved efficiency, protection, and dependability.Supercomputers supported complex scientific calculations; mainframes handled large-scale transactions and data analysis.
- Resource pooling: Cluster and commodity computing increased efficiency by coordinating multiple computers or nodes to execute a single operation inexpensively.Clusters require rapid LAN communication, and commodity clusters use open standards and ready-made processors.
- Connectivity and scale: Internet growth required scalable naming, networking, and web standards, including DNS, TCP/IP, HTTP, and HTML, to connect expanding hosts and services.DNS replaced overwhelmed centralized naming through hostname-to-IP translation, while web standards supported interconnected networks and services.
- Resource pooling: Grid computing coordinated groups toward shared objectives, especially non-interactive and IO-intensive tasks, while providing resource management and monitoring support.The model prioritizes applications running on a grid and adds reliability, tracking, and exploration capabilities.
- Abstraction and interoperability: Service-oriented, virtualized, and peer-to-peer models reduced dependence on bespoke infrastructure through reusable APIs, guest systems, and mediator-free resource sharing.These paradigms support interoperability, hardware abstraction, pooled virtual machines, and direct peer interaction.
4. Classification of Computing: Paradigms, Technologies and Trends
The paper classifies computing systems using three trait dimensions: their primary focus or paradigm, their technologies or impact areas, and their observed trends.
- Classification dimensions: Computing systems are classified by focus or paradigms, technologies or impact areas, and trends or observations.These traits provide the organizing dimensions for the broad categories presented in the classification.
4.1. Standalone vs. Networked Computing
The paper distinguishes standalone computing from networked computing and organizes each through paradigms, technologies or impact areas, and trends. Standalone systems emphasize embedded and compact hardware, while networked systems emphasize connectivity, cloud infrastructure, IoT, and distributed processing.
- Top-level distinction: Standalone computing uses computers without wired or WiFi connections, whereas linking multiple computers produces networked computing.The classification treats standalone and networked computing as distinct top-level categories.
- Standalone paradigms: Standalone paradigms include personal computers and embedded systems, ranging from user-operated microprocessor systems to computers integrated invisibly into devices.Examples include word-processing PCs, microwave controllers, and vehicle engine control units.
- Standalone technologies: Standalone technologies include single-board computers, Raspberry Pi platforms, and NVIDIA Jetson GPUs for compact, programmable, embedded, and robotic applications.Jetson systems support sensor fusion, visualization, obstacle detection, and path planning, with up to 32 TOPS of AI efficiency.
- Standalone trends: Standalone trends include adoption of low-power AI/ML systems for IoT services and cybersecurity protections spanning hardware, firmware, and applications.Embedded security addresses unauthorized use and fraudulent attacks on compact, resource-constrained devices.
- Networked paradigms: Networked computing includes connectivity infrastructure for cloud services, with SDN and NFV targeting adaptable, programmable networks and lower provider costs.Cloud networking must support communication within and between data centers while addressing scalability and addressing requirements.
- Networked trends and challenges: Networked systems combine cloud, fog, edge, and IoT layers for adaptive, near-real-time processing, while research remains focused on data safety, energy use, cost, and dependability.Fog processing can adapt sensor sampling after detecting a heart attack, and SDN-based cloud systems must balance security, cost, energy, and reliability.
4.2. General Purpose vs. Specialized Computing
General-purpose computing can support broad workloads, while specialized systems target intelligent applications and specific functions. The section surveys architectures, programming models, virtualization, data-management challenges, exascale systems, and large-scale machine learning.
- General-purpose computing uses CPU, memory, input/output, and a bus to support diverse computations, whereas special-purpose systems target intelligent applications.
- Von Neumann architecture connects CPU, memory, and I/O through a single bus, handling data and instructions in the same way.
- Programming Models: Programming models must balance efficiency, adaptability, target architecture, and code maintainability while coordinating sequential and parallel data-processing stages.
- Data Management: Big-data infrastructure must address data distributed across multiple systems by enabling communication among infrastructures, programming models, and execution settings.
- Virtualization: Virtualization decouples software and operating systems from physical hardware, while hypervisors let multiple operating systems share underlying resources.
- Virtualization: Cloud-based virtualization can streamline software construction through dynamic resource allocation, but differing provider APIs make migration and vendor-lock-in avoidance difficult.
- Exascale Computing: Exascale computing accelerates scientific knowledge generation by modeling complex physical processes, but its systems require substantial construction funding and cooling.
- Large-Scale Machine Learning: Large-scale machine-learning systems research emphasizes generally applicable, easy-to-develop methods supported by scientifically validated accuracy and scalability assurances.
4.3. Centralized vs. Decentralized Computing
Centralized computing relies on a central server to control and process most data, whereas decentralized computing has no single entity with overall influence.
- Centralized networks place most data control and processing in a central server, while decentralized networks lack a single controlling entity.
4.3.1. Centralized Computing
Centralized computing includes cloud-based paradigms and technologies that provide scalable services, storage, modular applications, containers, and serverless execution. Its development is shaped by networking, interoperability, energy, security, performance, and sustainability challenges.
- Cloud Computing: Cloud computing provides on-demand computing power, adjustable resource consumption, and usage-based payment, reshaping interactions with applications and systems.
- Cloud Computing: Mobile cloud computing combines mobile devices, cloud computing, and communication networks so handheld devices can run complex applications.
- Cloud Computing: Mobile cloud computing remains constrained by inadequate network capacity, service accessibility problems, security risks, cross-platform requirements, and battery drain.
- Green Cloud Computing: Green cloud computing targets lower energy consumption through workload and resource scheduling that accounts for carbon emissions.
- Green Cloud Computing: Micro-data centers and hardware-lifetime management are presented as approaches for improving efficiency, accommodating growth, dissipating heat, and reducing costs.
- Cloud Technologies: Cloud storage lets users access files through web or secure network connections while providers host, manage, maintain, and secure the underlying servers.
- Cloud Technologies: Microservices divide application functionality into autonomous services communicating through simple APIs, enabling components to be developed and scaled more independently.
- Cloud Technologies: Containers provide an inexpensive deployment environment by isolating applications, but shared hardware can cause performance interference and side-channel attacks.
4.3.2. Decentralized Computing
Decentralized computing encompasses paradigms that distribute computation, storage, and services across peers, edge locations, and other network layers. The section also highlights security, orchestration, and resource constraints affecting these systems.
- Focus or paradigms: Parallel computing divides work among processors so multiple tasks can execute simultaneously, supporting demanding IoT and big-data workloads.The section notes that conventional processors may be approaching practical limits for parallel calculations.
- Focus or paradigms: P2P networks distribute workloads and resource sharing among peers without a central mediator or server.Common uses include interactive media, file sharing, and content distribution.
- Focus or paradigms: Osmotic computing combines IoT, cloud, fog, and edge technologies to manage changing IT services and improve edge-cloud IoT operations.Its applications include urban planning, healthcare, connected vehicles, and Industry 4.0.
- Focus or paradigms: Dew computing supplements cloud services with locally operating resources, reducing dependence on continuous Internet access while offering low latency and user control.It is positioned as a bridge between cloud and on-premises computing rather than a replacement for cloud computing.
- Focus or paradigms: Edge computing places computation and storage near users, enabling real-time processing while reducing cloud transmission delays and backbone costs.This is especially relevant for AI applications involving high-definition video and real-time inputs, outputs, or retraining data.
- Technologies and impact areas: Blockchain-based distributed clouds can move data closer to users and provide on-demand resources, security, and potentially more cost-effective infrastructure.The section also identifies storage, scalability, privacy, and resource limitations as continuing challenges for blockchain in fog and edge systems.
4.3.3. Hybrid Computing
Hybrid computing combines centralized and decentralized networks to coordinate heterogeneous resources across cloud, fog, edge, IoT, and other distributed environments. The section presents decentralized intelligence and distributed operational objectives as responses to the limits of cloud-centric approaches.
- Focus or paradigms: Hybrid computing combines centralized and decentralized networks for systems that span multiple forms of infrastructure.The section frames hybrid computing through its paradigms, technologies, affected domains, and trends.
- Focus or paradigms: Fog-cloud-edge orchestration supports IoT applications that require heterogeneous distant, local, and multi-cloud resources alongside geographically distributed sensors.Examples include smart cities, automated manufacturing, virtual reality, and autonomous vehicles.
- Technologies and affected domains: Cryptocurrencies use decentralized networks and blockchain to record transactions without a centralized authority.Bitcoin is identified as the first decentralized digital currency and was launched in 2009.
- Technologies and affected domains: The machine economy describes exchanges of power, data storage, processing power, currency, and network connections among future computer networks.Cloud data centers, the web, and monetary exchanges are presented as components of this emerging network.
- Trends and observations: Distributed Computing Continuum Systems combine heterogeneous IoT, edge, fog, and cloud tiers to meet demanding computational requirements.Their geographically distributed and resource-constrained infrastructure requires approaches that adapt dynamically to application and user demands.
- Trends and observations: DCCS uses decentralized intelligence and distributed service-level objectives so components can make autonomous decisions based on local tasks and conditions.Modularizing and distributing SLOs is presented as a way to achieve scalable intelligence across the system.
4.4. Computational Methodologies: Parallel vs. Sequential Computing
The section contrasts parallel computing, which executes multiple tasks concurrently, with sequential computing, which processes operations in order. It presents parallelism as increasingly important as single-processor gains slow, while identifying efficiency, scalability, and programming challenges.
- Parallel computing: Parallel computing executes multiple tasks concurrently using several processors or threads, potentially reducing execution time.Parallel outputs are pooled after concurrent processes complete.
- Parallel computing: Multi-processor and multi-core systems divide large computations into smaller parts to improve execution speed, performance, and energy use.Multiple functional units can execute identical or distinct tasks in parallel.
- Parallel computing: ASICs optimize speed and energy efficiency for specific functions, whereas FPGAs can be reprogrammed to meet changing system requirements.FPGAs support both prototyping and later system versions, helping circumvent obsolescence.
- Parallel trends: Emerging parallel-computing trends include neuro-symbolic AI, scalable designs, and component-based systems that simplify monitoring, updating, troubleshooting, and scaling.Neuro-symbolic approaches are discussed in relation to the opacity and limited applicability of purely sub-symbolic methods.
- Sequential computing: Sequential computing executes operations in a specified order, with each operation’s output feeding the next operation’s input on a single processor.Traditional Von Neumann architecture follows this sequential instruction-processing model.
- Sequential trends: Energy efficiency, in-memory computing, and performance optimization are major sequential-computing concerns as processing capacity and single-processor efficiency face constraints.The section argues that parallelism must take a greater role while substantial programming and hardware-design obstacles remain.
4.5. Computing Trends and Emerging Technologies
Emerging computing trends target adaptability, AI capability, explainability, quantum technologies, and sustainability. The section also emphasizes that these advances face technical constraints and the environmental demands of expanding computing infrastructure.
- Advanced computing styles: Quantum AI is presented as a potentially transformative combination of quantum computing and artificial intelligence.The section links this possibility to broader effects on computing, AI, and the economy.
- Advanced computing styles: Quantum computing has produced algorithms for tasks considered difficult for conventional computers, but current technologies lack sufficient accuracy for reliable execution of some algorithms.The section specifically discusses precision requirements for initialization, multi-qubit operations, and quantum-state storage.
- Emerging technologies: Edge AI performs AI calculations near users on physical devices rather than concentrating all data processing in centralized data centers.Its applications are associated with IoT devices, edge computing, and uses such as medical diagnosis, driving, and agriculture.
- Emerging technologies: Bio-inspired computing draws on natural systems to create autonomous, flexible, self-learning, and reconfigurable computer architectures.The section also describes biological processes as subjects that computer science can model and understand.
- Emerging technologies: Explainable AI addresses whether automated systems can execute workloads reliably and whether their decisions can be understood.The section frames incorrect decisions in complex technologies as costly in resources and money.
- Industry and sustainability: Carbon-neutral computing is identified as a major challenge because expanding data-center infrastructure increases environmental scrutiny and requires progress in energy, water, and land efficiency.The section notes that each bit of data has a carbon footprint and that some regions lack access to green power.
5. Impact and Performance Criteria
Modern computing performance is assessed through QoS, SLA, autoscaling, fault tolerance, energy, carbon, serviceability, and human-centered criteria. The section highlights challenges in predicting behavior, managing resources, maintaining resilience, and limiting environmental costs.
- QoS and SLA: QoS metrics include execution time, cost, scalability, elasticity, latency, and dependability, while SLAs define standards and penalties.
- QoS and SLA: Unified SLA standards remain necessary across cloud providers to support multi-cloud operation.
- Autoscaling: Autoscaling dynamically modifies computational resources to reduce costs without compromising QoS, but latency-sensitive decisions must be made quickly.
- Fault Tolerance: Fault tolerance uses checkpointing and replication to preserve progress and maintain cloud-service reliability and availability.
- Efficiency: Data growth pushes cloud systems toward energy limits, encouraging fog and edge architectures while preserving QoS and reducing latency and cost.
- Efficiency: Computing research increasingly treats energy use and carbon footprints as system-level concerns involving power stability, clean energy, and alternate reduction strategies.
5.3. Social Impact
Social impact criteria include the digital divide, ethical AI, digital humanism, privacy, and resilience to attacks. Modern cloud, edge, fog, and IoT adoption creates both access opportunities and new security, privacy, and societal challenges.
- Digital Divide: The digital divide restricts rural businesses’ access to broadband, cloud resources, and affordable IT infrastructure.
- Ethical AI: AI systems require extensive business and client data, creating privacy, security, ownership, and regulatory-compliance challenges.
- Digital Humanism: Digital humanism is presented as a counter-force to business-driven digital colonization and the social tensions of global interdependence.
- Security and Compliance: Edge computing’s low latency, geographic dispersion, and end-device accessibility broaden confidentiality and information-safety concerns.
- Security and Compliance: Fog decentralization introduces security concerns absent from centralized cloud settings, including threats affecting authentication and IoT devices.
5.5. Economic and Management
Economic and management criteria cover cost-efficiency, resource allocation, application design, computing economics, and data management. The section connects growing IoT and cloud workloads with requirements for scalable, reliable, secure, and economical architectures.
- Cost-Efficiency: Cloud cost optimization balances application performance and efficacy against cloud expenditures, combining business management with technical analysis.
- Resource Allocation: Large-scale distributed architectures make resource management difficult because of diverse network devices, connections, servers, and infrastructure requirements.
- Resource Allocation: Kubernetes abstracts cluster-resource management so users can focus on composing cloud-native applications.
- Application Design: IoT systems require dependable, fast, scalable architectures that manage massive data volumes while preserving latency, security, and trustworthy communication.
- Application Design: Fog computing processes selected IoT data near devices, reducing cloud load and costs for geographically distributed, real-time applications.
- Computing Economics: Serverless computing bills resources only when functions are invoked, making suitability dependent on runtime behavior and workload.
- Data Management: Data lakes store unstructured cloud data but can be difficult to use because metadata handling and information organization are limited.
6. Emerging Trends in Modern Computing
The review classifies modern computing developments by maturity and presents them through a hype cycle. Edge AI and federated learning are prominent current research areas, while several other paradigms remain earlier-stage prospects.
- Maturity Levels: The analysis groups computing areas into maturity horizons of under five years, five to ten years, and over a decade.
- Current Trends: Edge AI and federated learning currently dominate research attention among emerging computing areas.
- Emerging Areas: Distributed computing continuum and AI-driven computing are described as emerging areas whose research is still developing.
- Longer-Term Prospects: Biologically inspired, neuro-symbolic, analog, neuromorphic, 6G, and quantum computing are not yet considered hype-worthy in the review.
- Longer-Term Prospects: Cloud and fog computing have trended heavily and may continue trending for the next five to ten years.
7. Summary and Conclusions
The research reviews modern computing’s evolution over six decades, from classical systems to quantum computing, while examining architectures, conceptual units, and communication methods. It also connects shifts toward decentralized and specialized computing with IoT, AI, accessibility, and societal impact.
- The article traces computing’s transition from classical computers to quantum computing across the past sixty years.
- It examines how physical architecture, conceptual units, and communication methods shape modern computing systems.
- Changes in microcontroller architecture, operating systems, and networking infrastructure enabled ubiquitous models such as the Internet of Things.
- The study identifies specialized hardware and software for data-driven fields such as AI as a shift from earlier emphasis on flexibility and adaptability.
- It addresses accessibility and potential inequalities, emphasizing that computing technologies should benefit society and everyday life.
Appendix A. List of Acronyms
Appendix A presents Table 6 as a list of acronyms.
- Table 6 is identified as the list of acronyms.
- The appendix directs readers to Table 6 for acronym information.
- The acronym material is presented as an appendix table.