PaperlayerView exact revision on arXiv

Source-linked AI summary

A Manifesto for Future Generation Cloud Computing: Research Directions for the Next Decade

Rajkumar Buyya, Satish Narayana Srirama, Giuliano Casale, Rodrigo Calheiros, Yogesh Simmhan, Blesson Varghese, Erol Gelenbe, Bahman Javadi, Luis Miguel Vaquero, Marco A. S. Netto, Adel Nadjaran Toosi, Maria Alejandra Rodriguez, Ignacio M. Llorente, Sabrina De Capitani di Vimercati, Pierangela Samarati, Dejan Milojicic, Carlos Varela, Rami Bahsoon, Marcos Dias de Assuncao, Omer Rana, Wanlei Zhou, Hai Jin, Wolfgang Gentzsch, Albert Y. Zomaya, Haiying Shen

arXiv:1711.09123v2cs.DC

TL;DR

Cloud computing’s expanding adoption and emerging models create open challenges involving security, privacy, sustainability, networking, and economics. This manifesto synthesizes current advances and limitations, identifies emerging trends and impact areas, and proposes research directions for the next decade. It concludes that these directions offer a promising future for Cloud computing technically and economically while calling for community action.

  • Problem

    Expanding Cloud adoption and emerging models create unresolved challenges in areas including security, privacy, sustainability, networking, and economics.

  • Method

    The manifesto brings together advancements, summarizes the state of the art and its limitations, discusses emerging trends and impact areas, and proposes future research directions.

  • Results

    The manifesto identifies major open Cloud computing challenges and offers comprehensive research directions for the next decade.

  • Takeaways & Limitations

    The proposed research directions indicate a promising technical and economic future for Cloud computing and call the community to address the identified open issues.

  • Takeaways & Limitations

    Practical privacy-preserving data-access solutions remain an open problem, including protection against access-frequency attacks and information leakage.

Abstract

from arXiv · show

The Cloud computing paradigm has revolutionised the computer science horizon during the past decade and has enabled the emergence of computing as the fifth utility. It has captured significant attention of academia, industries, and government bodies. Now, it has emerged as the backbone of modern economy by offering subscription-based services anytime, anywhere following a pay-as-you-go model. This has instigated (1) shorter establishment times for start-ups, (2) creation of scalable global enterprise applications, (3) better cost-to-value associativity for scientific and high performance computing applications, and (4) different invocation/execution models for pervasive and ubiquitous applications. The recent technological developments and paradigms such as serverless computing, software-defined networking, Internet of Things, and processing at network edge are creating new opportunities for Cloud computing. However, they are also posing several new challenges and creating the need for new approaches and research strategies, as well as the re-evaluation of the models that were developed to address issues such as scalability, elasticity, reliability, security, sustainability, and application models. The proposed manifesto addresses them by identifying the major open challenges in Cloud computing, emerging trends, and impact areas. It then offers research directions for the next decade, thus helping in the realisation of Future Generation Cloud Computing.

1 Introduction

Cloud computing has become foundational to modern IT through subscription-based services, while new models and geographically distributed infrastructure create further opportunities and challenges. The manifesto surveys these developments, identifies open issues, and proposes research directions for the next decade.

  • Cloud computing has become the backbone of the modern economy by providing on-demand, subscription-based IT resources.
  • SaaS, PaaS, and IaaS have been the three main service models fostering Cloud adoption.SaaS provides the highest abstraction by delivering hosted applications over the Internet.
  • Containers, specialised Cloud services, and serverless architectures extend Cloud capabilities beyond traditional virtualisation-based models.Containers contributed to the emergence of Container as a Service, while analytics services and serverless architectures support specialised business requirements.
  • Geographically distributed Clouds are extending infrastructure toward edge devices to support IoT and Fog computing.
  • Cloud adoption and spending were projected to grow substantially, increasing pressure to rethink research strategies and existing models.The cited 2016 reports found 70% of organisations had at least one application in the Cloud and projected spending above $203 billion in 2020.
  • The manifesto identifies challenges, reviews state-of-the-art solutions and limitations, examines emerging trends, and proposes future research directions for the next decade.Its scope targets impacts across the next two or three generations of utility-oriented Cloud technologies, infrastructures, and application services.

2 Challenges: State-of-the-Art and Open Issues

Cloud computing’s open issues span scalability, reliability, sustainability, heterogeneity, interoperability, privacy, security, and migration. Existing approaches have advanced these areas but remain limited by system complexity, inaccurate workload information, energy demands, incompatible interfaces, privacy leakage, and incomplete provider-selection and migration frameworks.

  • Scalability and Elasticity: Cloud scalability is constrained by the end of Moore’s and Dennard scaling, requiring heterogeneous hardware abstractions that preserve isolation and performance guarantees.The challenge spans hardware, middleware, and application levels, including multi-core systems, clusters, GPUs, neuromorphic architectures, and quantum architectures.
  • Scalability and Elasticity: Resource-management policies remain vulnerable to inaccurate workload estimates, leaving unresolved trade-offs between policy optimality and robustness.Demand estimation and workload prediction methods are described as brittle, and the role of ML and AI remains open.
  • Reliability: Cloud reliability remains difficult to model because service-oriented, massive-scale systems combine interdependent failures across heterogeneous hardware, software, and wide-area networks.Existing environments lack thorough service reliability models, automatic reliability-aware management, and failure-aware provisioning policies.
  • Sustainability: Sustainability requires optimization beyond server VM consolidation because cooling and high-speed networks also consume substantial data-centre energy.Cooling systems account for close to 35% of energy consumption, while ICT overall uses close to 10% of global electricity.
  • Heterogeneity: Heterogeneous Cloud resources make accelerator-aware software difficult because vendor-specific languages and low-level programming requirements do not align easily with high-level Cloud abstractions.Middleware abstraction can reduce source-code optimization opportunities, while infrastructure services leave developers responsible for targeting hardware environments.
  • Interconnected Clouds: Cloud interoperation remains open because providers use incompatible proprietary APIs, formats, and contextualization mechanisms, while interoperability also requires portable data, applications, interfaces, and recognized quality and security standards.Existing middleware integrations may operate only at the lowest common denominator, producing suboptimal applications or limited service-model support.
  • Security and Privacy: Cross-domain Cloud applications face privacy risks because data mining can reveal more personal information than anticipated, while access-frequency attacks remain insufficiently addressed by practical privacy-preserving indexing.The passage identifies practical solutions based on PIR, Oblivious RAM, B-trees, and binary search trees as an open problem.
  • Security and Privacy: Cloud security against Advanced Persistent Threats requires both technical-driven and policy-driven solutions because such attacks can disable conventional defensive mechanisms.APTs are characterized as targeted, organized, funded, technically advanced, stealthy, and persistent, affecting data and computation confidentiality, integrity, and availability.

3 Emerging Trends and Impact Areas

Emerging Cloud trends—including containers, Fog, serverless computing, SDN/NFV, and blockchain—expand deployment and management possibilities while introducing security, privacy, resource, and software-engineering challenges.

  • 3.1 Containers: Containers provide lightweight, isolated application environments and have driven the emergence of Container as a Service.They package software and dependencies together while sharing a compute node’s resources through isolated user-space processes.
  • 3.1 Containers: Less than a second startup and small resource footprints make containers faster and more resource-efficient than VMs, supporting more simultaneous instances.The passage explicitly attributes container popularity to rapid launching and low memory consumption.
  • 3.1 Containers: Container adoption remains constrained by weaker kernel-sharing isolation, startup optimization needs, and QoS-aware cluster management.Proposed directions include trusted execution hardware, Unikernels, storage-driver optimization, and systems such as Kubernetes, Mesos, and Swarm.
  • 3.2 Fog Computing: Fog computing extends Cloud resources to network edges, reducing latency and supporting location-aware, mobile, streaming, real-time, and IoT applications.Its ecosystem includes nodes such as mobile base stations, gateways, and routers that process data near its source.
  • 3.2 Fog Computing: Fog computing requires interoperable multi-party SLA management alongside stronger security and privacy across interactions among users, edge nodes, and Cloud data centres.Different organizations may own edge nodes, complicating responsibility assignment and unified management.
  • 3.3 Serverless Computing: Serverless functions create resource-management and software-engineering challenges because arbitrary code executes under provider restrictions and variable performance conditions.Applications may need redesign, including statelessness, while function execution time and capabilities are constrained.
  • 3.4 Software-Defined Networking and Network Function Virtualization: SDN/NFV enable programmable network control and software-based network functions, supporting more agile, elastic, and on-demand Cloud networking.Their combination opens research directions for sustainability, interconnected Clouds, security, service-chain placement, and resource management.
  • 3.5 Blockchain: Blockchain can provide more secure and auditable Cloud transactions for collaboration and data exchange across sectors such as health, agriculture, manufacturing, and petroleum.The passage connects this role to the importance of data for machine-learning and deep-learning solutions.

4 Future Research Directions

Future Cloud research must address new requirements from heterogeneous IoT data, personalized services, and evolving application demands by advancing scalability, elasticity, infrastructure reuse, deployment, reliability, and InterCloud execution.

  • 4 Future Research Directions: The next decade will require Cloud services that process large heterogeneous IoT data streams while adapting to diverse hardware, QoS constraints, energy use, and costs.The stated requirements include energy- and cost-aware personalization and optimization across multiple criteria.
  • 4 Future Research Directions: The manifesto reviews existing research and limitations, then proposes future Cloud research directions shaped by emerging technological trends.It targets the coming decade and anticipates impact across the next two or three generations of Cloud technologies and services.
  • 4.1 Scalability and Elasticity: Special-purpose Clouds for deep learning, data-stream analytics, and image or video recognition are identified as hardware-level scalability and elasticity directions.Examples include CNNs, MLPs, and LSTMs, whose narrow functions can support a broad application spectrum.
  • 4.1 Scalability and Elasticity: Middleware research should improve infrastructure reuse, accelerate large-scale deployment and provisioning, support geographically remote failovers, and enable computation across multiple public Clouds.HPC settings additionally require consistent performance and scheduling that avoids execution noise or uses dedicated clusters.
  • 4.1 Scalability and Elasticity: Large communication-intensive applications require high-throughput, low-latency networks and mechanisms that support multiple clients while maintaining sustainable and profitable provision.Big Data applications increasingly leverage HPC and IoT capabilities for applications such as smart cities.

4.2 Resource Management and Scheduling

Resource management and scheduling must adapt to decentralized, heterogeneous Cloud–Edge environments and serverless execution while balancing performance, QoS, cost, privacy, security, energy, and provider–consumer objectives.

  • 4.2 Resource Management and Scheduling: Fog, Edge, serverless, software-defined computing, and Big Data trends increase decentralization, resource heterogeneity, workload variability, and offloading decisions.These trends create new questions about how much application logic should move away from the system core.
  • 4.2 Resource Management and Scheduling: Edge analytics requires cooperative management across centralized Cloud data centres and distributed Edge resources for real-time processing of stream-based or event-driven sensor data.The passage places processing across the complete Fog-topology hierarchy.
  • 4.2 Resource Management and Scheduling: Scheduling methods such as multi-resource fairness and game-theoretic approaches must be generalized to large-scale heterogeneous environments spanning Cloud data centres and Edge nodes.Mean-field games are proposed for reasoning about interactions among many resources, devices, and user types.
  • 4.2 Resource Management and Scheduling: FaaS management must account for erratic execution performance, network latency, per-access pricing, QoS expectations, privacy, and security when deciding whether to offload computation.The provider must optimize resource allocation while the user requires suitable execution quality.
  • 4.2 Resource Management and Scheduling: FaaS providers need proactive provisioning and service-class allocation policies that balance resource efficiency, latency, revenue, and user QoS.Different provider and consumer goals motivate game-theoretic and market-oriented allocation models.
  • 4.2 Resource Management and Scheduling: SDN/NFV enable holistic resource control, joint VM–traffic consolidation, and on-demand service-chain placement, but require new placement heuristics and management policies.These capabilities can target energy consumption, SLA satisfaction, and dynamic deployment of network services.
  • 4.2 Resource Management and Scheduling: Machine learning and AI may address complexity, heterogeneity, and scale in provisioning, consolidation, and load balancing, while potentially reducing control and determinism.Explainable AI is identified as a direction for facilitating industry adoption of AI-based Cloud management.
  • 4.2 Resource Management and Scheduling: Extreme-scale scientific workflows require fault tolerance, performance modeling, efficient data management, and efficient resource use as processed data volumes increase.Big Data analytics is presented as a crucial tool for addressing these concerns.

4.3 Reliability

Reliability research must address failures across Cloud and Fog systems while balancing performance, QoS, energy use, and storage trade-offs. Proposed directions include intelligent failure prediction, fault-tolerant storage, and adaptive online management.

  • Reliability challenge: Cloud providers must meet performance, reliability, and QoS expectations despite independent and correlated failures.Reliability directly affects QoS and providers’ long-term reputation.
  • Failure prediction: Deep and machine learning can support failure prediction and failure-aware resource provisioning for reliability and performance guarantees.The approach depends on characterising failures from massive failure datasets.
  • Fog reliability: Reliability-aware Fog approaches must also account for power efficiency because replication and intelligent fault handling can create energy trade-offs.The challenge is especially difficult for applications distributed across numerous edge components.
  • Storage reliability: Cloud storage research must improve fault tolerance for Big Data while balancing durability, availability, storage overhead, bandwidth, energy consumption, and recovery performance.Replication and erasure coding provide alternative techniques with different trade-offs.
  • Adaptive management: Online management must consider QoS, SLAs, and energy consumption simultaneously because fast-changing behaviours cannot be completely predicted in advance.The paper calls for adaptive, self-aware techniques.

4.5 Heterogeneity

Cloud heterogeneity remains limited by gaps between unified management and diverse resources, and between abstraction and accelerator-specific programming. Future directions target integrated management and higher-level portable programming models.

  • Unified management: Heterogeneity lacks a unified environment spanning VM, vendor, and hardware-architecture levels of resource and workload management.The manifesto proposes an umbrella platform that selects strategies according to application requirements.
  • Unified management: Such a platform requires heterogeneous memory management beyond hypervisor-centered solutions, which limit heterogeneity’s benefits.The proposed direction integrates a portfolio of workload and resource-management techniques.
  • Programming abstraction: Accelerator programming remains difficult because current models require accelerator-specific languages and low-level programming effort.This restricts broader use of heterogeneous hardware for service-oriented and user-driven Cloud applications.
  • Programming abstraction: Open-source high-level programming models are needed to bridge abstraction and heterogeneity as applications use both Cloud and Fog resources.The paper identifies this direction as relevant to the emerging Fog ecosystem.
  • Programming abstraction: OpenCL offers limited performance portability, while higher-level approaches such as TANGRAM still need performance models and adaptive runtimes.Automation is also needed for fine-grained task partitioning and CPU–accelerator interaction.

4.6 Interconnected Clouds

Interconnected Clouds require more than standardisation: they need composable middleware, user- and provider-facing interfaces, economic incentives, and network technologies supporting interoperability, security, and QoS.

  • Interoperability foundations: Interoperability and portability across Cloud systems remain highly complicated, and pure standardisation is insufficient.Containers, configuration tools, adapters, and libraries are practical but incomplete approaches.
  • Service composition: Cloud middleware should compose complex functionalities from services offered by one or multiple providers without forcing a minimum common functionality set.This preserves access to provider-specific capabilities.
  • Service composition: User-level InterCloud middleware must identify composition candidates despite providers’ incentives to retain customers and functionalities.The challenge concerns service discovery without provider support.
  • Middleware interfaces: Composition middleware needs seamless user-facing delivery and provider-facing mechanisms for combining or accessing services across providers.The delivered functionality may come from one provider or from multiple composed services.
  • Cloud federation economics: Cloud federation depends on vendors seeing financial benefits, motivating research on billing, accounting, pricing, and InterCloud marketplaces.Economic incentives are presented as a condition for ubiquitous interconnection.
  • Network support: SDN and NFV warrant investigation for addressing interoperation challenges, including improved security and QoS.SDN has already been used to optimise wide-area traffic between data centres.

4.7 Empowering Resource-Constrained Devices

Resource-constrained devices face adaptability, incentive, and underexplored technology challenges as mobile Cloud, IoT, Fog, Edge, and container-based execution evolve. Edge analytics is proposed to reduce network latency by processing sensor data across the Fog hierarchy.

  • Mobile Cloud: Mobile Cloud code offloading still has adaptability issues, while increasing smartphone capabilities limit the applications that benefit from offloading.Task delegation is established, but offloading opportunities are becoming narrower.
  • Incentives: Incentive mechanisms are needed to encourage participation in heterogeneous mobile Cloud offloading and reward users appropriately.The paper also connects these mechanisms to social networking, IoT, and Fog domains.
  • Emerging technologies: The benefits of serverless computing, CaaS, and Fog computing for mobile Cloud applications remain insufficiently explored.These technologies are identified as emerging opportunities for the domain.
  • Containers at the Edge: Lightweight containers are expected to become widely deployed on edge servers for application packaging, runtime support, and offloading.Customised containers such as Cloud Android Container target Edge computing and offloading features.
  • Edge analytics: Edge analytics processes IoT sensor data across the Fog hierarchy rather than sending all data to the Cloud, addressing network-latency concerns.The direction targets real-time streaming data analytics.

4.8 Security and Privacy

Future Cloud security and privacy research must address diverse trust assumptions, protection requirements, distributed infrastructures, and emerging Fog environments. Key directions include scalable encrypted data access, selective sharing, infrastructure protection, and Cloud-specific DDoS handling.

  • Trust assumptions: Cloud security solutions must match scenario-specific trust assumptions, ranging from completely untrusted providers to trustworthy ones.Security and privacy concerns span networks, databases, virtualization, and resource scheduling.
  • Data protection: Client-side encryption requires scalable techniques that preserve functionality, integrate with Cloud technology, limit information leakage, and support rich queries.Selective retrieval indexes and query-supporting encryption can themselves create leakage risks.
  • Data protection: Multi-provider data protection should investigate data splitting for confidentiality and integrity proofs across independent Cloud providers.The research direction also covers generic confidentiality and user-defined access or visibility constraints.
  • Data sharing: Selective data sharing must support write privileges, multiple writers, and efficient policy updates across distributed storage involving independent providers.These requirements arise in scenarios with multiple users and possible multiple providers.
  • Fog security: Fog computing increases exposure to privacy breaches because decentralized environments combine trusted and malicious nodes without central controls.This motivates adapting secure routing, redundant routing, and trust-topology research.
  • Infrastructure security: Cloud infrastructure research must secure multi-tenancy, virtual-machine lifecycle and placement, and legitimate-request identification while limiting attack-related resource waste.Cloud systems should distinguish attacks such as excessive load and decide whether to drop it or avoid overprovisioning.

4.9 Economics of Cloud Computing

Cloud economics is being reshaped by containers, serverless computing, Fog and Edge resources, expanding marketplaces, and uncertain service environments. Research must develop viable incentives, business models, and architectures that balance value, dependability, and risk.

  • Emerging models: Lower container overheads can support real-time workloads, while serverless computing is becoming commonplace across major Cloud platforms.Examples include Google Cloud Functions, Amazon Lambda, Microsoft Azure Functions, and IBM Bluemix OpenWhisk.
  • Fog and Edge economics: Fog and Edge resources alongside Cloud data centres can reduce operating costs but may require incentives for users to make devices available.Their availability creates scope for additional business models and provider categories.
  • Fog and Edge economics: Micro Data Centres (MDCs) are positioned between traditional data centres and user-owned resources, enabling models such as dynamic discovery and pre-agreed contracts.Dynamic discovery can select providers by availability profile, security credentials, or type, but does not directly guarantee user objectives.
  • Uncertainty: Unpredictable Fog and Edge resources and dynamic provisioning require Cloud architects to embrace uncertainty in architectures and market models.Game-design strategies may dynamically manage Cloud tenants, service providers, and other participants.
  • Uncertainty: Future architectures and markets should seek continuous win-win resolutions for providers, users, and intermediaries while preserving value and dependability.The proposed direction explicitly treats these outcomes as a research objective under uncertainty.
  • Migration: Migrating in-house IT to the Cloud leaves longer-term effects on risk tolerance and business continuity unclear despite potentially greater external expertise.The discussion contrasts outsourcing benefits in cybersecurity, updates, and availability with unresolved organizational risks.
  • Marketplace: An expanding Cloud marketplace, including edge-focused opportunities for telecommunications operators, creates room for new providers alongside dominant incumbents.Recent emphasis on Edge and serverless computing supports this potential expansion.

4.10 Application Development and Delivery

Cloud application development must evolve beyond continuous delivery automation toward methods, architectures, patterns, and programming models that preserve quality across decomposed, adaptive, and multi-paradigm systems. Research priorities include infrastructure-as-code engineering, serverless integration, workload formalization, and orchestration trade-offs.

  • Application development: Continuous delivery can reduce design-time reasoning about SLA compliance, business alignment, and value-driven design, risking unsuitable early Cloud architectures.Research should support continuous monitoring and iterative evolution of application design and quality within delivery pipelines.
  • Application development: Infrastructure-as-code adoption is expected to grow, but software engineering methods tailored to writing, debugging, and evolving it remain insufficient.The challenge is complicated by infrastructure code combining different programming languages and technologies.
  • Architectures and patterns: Cloud-native architectures and design patterns should make application definition more human-oriented while tolerating changes in runtime context, technology, and service-level agreements.These styles must account for the runtime domain and changing operating conditions.
  • Architectures and patterns: Microservices and containers enable more application decompositions, with consequences for security, performance, reliability, and operational costs.Application decomposition is identified as a central architectural challenge.
  • Architectures and patterns: Serverless and FaaS require integration and control patterns that combine external services with serverless services and connect Edge cyber-physical systems to the Cloud.Cloudlets and swarmlets are proposed as application fragments supporting local decisions near sensors, actuators, and control layers.
  • Programming models: Integrated paradigms require workload formalisms that propagate performance, reliability, and security requirements through architectures using technologies such as blockchain, SDN, Spark, and Storm.These formalisms should characterize both execution requirements and properties of processed data.
  • Orchestration: Orchestration of integrated services must examine resource choices and trade-offs across security, privacy, and other quality dimensions.The passage contrasts easier protection of individual functions with the greater attack surface of FaaS-based architectures.
  • Programming models: Fog, Edge, InterCloud, and IoT applications need adaptive elastic decentralized programming models with modular specification, testing, and verification.Program specifications may be probabilistic when analyzing asynchronous data streams.

4.11 Data Management

Future Cloud data management must address provenance, latency, privacy, hybrid placement, and the specialized demands of IoT, deep learning, and blockchain. Research spans edge-aware services, auditable protection, high-level abstractions, and compact model and dataset storage.

  • Data provenance: Data Lakes need metadata tracking for dataset source and provenance because Cloud storage alone does not provide sufficient visibility into data origins.The passage contrasts this gap with structured data warehouses.
  • Edge data services: Centralized Cloud data centres provide economies of scale, but latency and global bandwidth limits motivate data services at the Fog layer.Fog data services are described as essential as IoT and 5G mobile networks spread.
  • Privacy and compliance: Regulatory support does not resolve proprietary-data concerns because enterprises still lack clear audit mechanisms showing what Cloud providers can access.The issue remains especially significant for sensitive trade secrets.
  • Hybrid and stream data: Hybrid Cloud, SDN, Fog elasticity, and flow routing require joint optimization of data placement, processing, and adaptation across network devices.Optimal placement of data-processing applications and stream-processing tasks remains difficult.
  • Big Data platforms: High-level programming abstractions can ease development and deployment of Big Data applications across hybrid infrastructures and commercial or open-source Cloud platforms.Platform bindings support deployment on providers including Google Cloud Engine and Amazon Web Services.
  • Emerging workloads: IoT data management must support streaming, efficient storage, seamless edge–Cloud integration, and trust and provenance for unmanaged edge devices.These requirements are part of specialized services for IoT, deep learning, and blockchain.
  • Emerging workloads: Deep learning requires efficient trained-model management and rapid loading or switching, while video and image training datasets demand compact storage techniques.Large critical datasets make training particularly costly for applications such as autonomous vehicles and augmented reality.
  • Emerging workloads: Cloud-hosted blockchains and distributed ledgers can provide data tracking, increased assurance, provenance, and implicit audit trails for enterprise records.Financial transactions are a leading use case, but the approach can extend to other enterprise data.

4.12 Networking

Future Cloud networking research must improve programmable traffic management, security, NFV orchestration, and capacity planning for increasingly distributed and demanding workloads.

  • Traffic engineering: SDN-based traffic engineering can use global network visibility for dynamic flow scheduling and management across Cloud data-centre networks.Future techniques are expected to target delay, bandwidth, and jitter guarantees for application QoS and committed SLAs.
  • Security and privacy: SDN may improve Cloud security and reliability, but its controllers and inter-SDN communications introduce new attack surfaces.The paradigm shift can also magnify the impact of existing threats.
  • NFV orchestration: NFV research should address VNF placement, consolidation, and coordinated auto-scaling while meeting application QoS and SLA requirements.Traffic engineering combined with VNF migration and placement may reduce network communication cost.
  • Network automation: AI, ML, and Big Data analytics, combined with SDN visibility, open opportunities for automating next-generation Cloud networks.The text notes that incorporation of these approaches remains unclear.
  • Capacity and edge computing: IoT and 5G will intensify network-capacity demands, while Edge/Fog computing reduces latency and resource usage but cannot address all networking demand.The transition therefore requires new network products and services.

4.13 Usability

Cloud usability research should help users estimate costs, understand complex resource choices, and evaluate systems across heterogeneous user populations.

  • Cost awareness: Workload and resource fluctuations make Cloud spending difficult to predict, motivating tools that improve cost estimation and user satisfaction.The passage also identifies visualization technologies as a way to understand infrastructure.
  • Resource selection: Advisory systems could reduce the burden of choosing among CPUs, GPUs, network, storage, operating systems, and PaaS services.They may also recommend more efficient Cloud usage.
  • Evaluation: Cloud usability evaluation is difficult because quantitative experiments and qualitative user studies address different needs across a heterogeneous user community.Users differ in profiles, needs, skills, and the Cloud layers in which they work.

4.14 Discussion

Future research spans all Cloud service models, from heterogeneous and non-traditional IaaS hardware to resource-efficient PaaS and more adaptable SaaS delivery.

  • IaaS: IaaS research will expand toward heterogeneous hardware, special-purpose Clouds, non-traditional architectures, containers, SDN, and Fog/Edge computing.Examples include GPUs, TPUs, neuromorphic systems, quantum computing, adiabatic computing, and nanocomputing.
  • PaaS: PaaS research emphasizes resource management, scalable elastic computing, heterogeneous-resource use, energy efficiency, portability, reliability, and performance.The passage frames these goals as outcomes of proposed programming abstractions, models, languages, and systems.
  • Data management and analytics: Data management and analytics research includes platform support for edge analytics and real-time stream processing in IoT and smart-city domains.The discussion places these directions alongside security research.
  • SaaS: SaaS research should simplify application development and delivery while supporting continuous monitoring, iterative evolution, and emerging technologies.Target technologies include IoT, Edge computing, Big Data, and serverless computing.
  • Cloud economics: Cloud architectures and market models should accommodate uncertainty and pursue continuous win-win resolutions for providers, users, and intermediaries.The supported objectives combine return on investment with satisfaction of SLA perspectives.

5 Summary and Conclusions

The manifesto argues that emerging requirements will challenge existing Cloud models and presents research directions for addressing open issues over the next decade.

  • Emerging requirements: Cloud’s next decade will require energy- and cost-aware personalised services that adapt across heterogeneous hardware while meeting QoS and economic constraints.Large-scale IoT and sensor networks will also produce substantial data streams to store, manage, and analyse.
  • Open challenges: These requirements call for new research approaches and re-evaluation of models addressing scalability, resource provisioning, and security.The text presents this as a response to the challenges created by changing Cloud workloads and technologies.
  • Manifesto contribution: The manifesto synthesises current challenges, state-of-the-art limitations, emerging trends, impact areas, and future research directions for the next decade.It concludes by calling the Cloud community to address these open issues.
Loading 1711.09123v2…