Source-linked AI summary
Research Agenda in Cloud Technologies
Ilango Sriram, Ali Khajeh-Hosseini
TL;DR
Cloud computing’s rapid industry development outpaced systematic academic synthesis of its technical research agenda. The paper systematically reviews peer-reviewed literature across major research databases and organizes findings around related technologies, standards and interfaces, cloud construction, and use cases. It finds strong emphasis on standards, interoperability, and cloud design, while noting that the fast-moving field makes the review quickly outdated.
Problem
Academia lacked a systematic review of the rapidly expanding peer-reviewed research agenda in cloud computing.
Method
The authors systematically searched major computer-science databases, screened more than 150 papers, and retained primarily peer-reviewed research relevant to cloud computing.
Results
The reviewed research concentrated on standards, interoperability, and methods for designing and building clouds, with emerging cloud use cases also examined.
Takeaways & Limitations
The review provides a research-agenda baseline that places cloud interoperability and related technical directions in context for researchers.
Takeaways & Limitations
Because cloud computing is a fast-moving field, the review will soon become out of date as new developments emerge.
Abstract
from arXiv · showhide
Cloud computing is the latest effort in delivering computing resources as a service. It represents a shift away from computing as a product that is purchased, to computing as a service that is delivered to consumers over the internet from large-scale data centres - or "clouds". Whilst cloud computing is gaining growing popularity in the IT industry, academia appeared to be lagging behind the rapid developments in this field. This paper is the first systematic review of peer-reviewed academic research published in this field, and aims to provide an overview of the swiftly developing advances in the technical foundations of cloud computing and their research efforts. Structured along the technical aspects on the cloud agenda, we discuss lessons from related technologies; advances in the introduction of protocols, interfaces, and standards; techniques for modelling and building clouds; and new use-cases arising through cloud computing.
1. INTRODUCTION
Cloud computing’s rapid industry growth prompted a systematic review because academia had only recently begun producing substantial peer-reviewed research. The paper surveys technical advances and the research agenda pursued by academia.
- The review was motivated by cloud computing’s rapid rise in IT and academia’s recent uptake of the topic.
- Earlier whitepapers and introductions existed, but no systematic academic review adequately covered the field’s research agenda and challenges.
- The paper reviews technical issues in protocols, interoperability, and cloud construction, while enterprise and user challenges are discussed elsewhere.
- The research agenda is divided between building and providing clouds and studying cloud computing’s implications for enterprises and users.
2. METHODOLOGY
The authors conducted a systematic literature search across major computer-science databases, then filtered the results to construct a peer-reviewed review corpus. The search covered cloud-related research published from 2005 through October 2009.
- The review searched ACM Digital Library, IEEE Xplore, SpringerLink, ScienceDirect, and Google Scholar using cloud-computing service and deployment terms.
- The search period was limited to 2005 through October 2009 because the survey began in October 2009.
- More than 150 papers were retrieved, after which titles and abstracts were screened for relevance and quality.
- The review retained peer-reviewed work, with a small number of relevant non-peer-reviewed definitions or workshop summaries, and excluded misleading, HPC-focused, and e-Science-focused papers.
3. DEFINITIONS
Cloud computing is framed as shared, virtualized resources delivered on demand and measured as a service. The paper distinguishes five essential characteristics, three service models, and four deployment models, while noting that definitions may evolve.
- The NIST definition describes cloud computing through five characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity, and measured service.
- SaaS provides browser-based access to provider-hosted software, while PaaS supports application development through provider-supported languages and tools.
- IaaS provides processing, memory, and storage through virtual machines, offering more flexibility than PaaS while making users responsible for operating-system maintenance.
- The three service models can be deployed on any of the four deployment models, separating service abstraction from cloud ownership and use.
- Private, public, community, and hybrid clouds classify deployment by ownership and use; hybrid clouds permit applications and data to move across component clouds.
- Cloud resources are pooled among users, can scale with variable loads, and are commonly delivered through pay-per-use arrangements with SLA guarantees.
4. Lessons from related Technologies
Cloud research draws on distributed, grid, SOA, and market-oriented computing while addressing cloud-specific scale, abstraction, business, and resource-management issues. Related work supplies reusable ideas but also exposes limits of cloud systems for some workloads.
- Lessons from related Technologies: Grid and cloud computing share a utility-computing vision, but differ in business models, security, resource management, and abstraction.
- Lessons from related Technologies: Market-oriented resource allocation uses brokers to regulate supply and demand, prioritizing service requests differently when resources are limited.
- Lessons from related Technologies: SOA contributes service principles and monitoring experience for robust distributed applications built from unreliable components.
- Lessons from related Technologies: In the LINPACK comparison, cloud nodes matched HPC nodes individually but suffered severe multi-node performance loss, preventing competition with Top500 supercomputers for scientific computation.
- Lessons from related Technologies: Cloud research emphasizes scalability, including decoupling to avoid synchronization and managing many low-end or faulty components.
- Lessons from related Technologies: The cloud community can learn from related fields, including functionality modelling from service computing and context sensitivity from pervasive computing.
5. STANDARDS AND INTERFACES
Cloud standards and interfaces are needed to reduce vendor lock-in and support interoperable cloud markets. Progress includes compatibility efforts, open-source implementation tools, and standardized APIs, but interfaces alone are insufficient for full interoperability.
- Open standards are needed to enable an open market for providing and consuming cloud resources, because vendor-specific solutions increase lock-in and trust barriers.
- Interoperability between IaaS providers requires machine-image, contextualization, and API-level compatibility.
- OVF is being developed as an open standard for machine-image or virtual-machine compatibility across providers.
- Eucalyptus emulates Amazon EC2 interfaces, allowing IaaS clouds built from clusters to use EC2 tools and providing an open-source platform for provisioning research.
- Standardized APIs could enable generic management tools across cloud offerings, but protocols and software artifacts are also needed to achieve interoperability and unlock broader benefits.
6. CLOUD INTEROPERABILITY AND NOVEL PROTOCOLS
Cloud interoperability research spans protocols, virtual-machine portability, SaaS integration, and abstraction layers across public and private clouds. Although a broad intercloud agenda has been defined, most communication and resource-sharing work still lacks initial empirical results.
- Intercloud protocols target interoperability through addressing, naming, identity and trust, presence and messaging, virtual machines, multicast, time synchronization, and reliable application transport.
- Virtual machine contracts aim to standardize VM protection and security settings as extensions to OVF metadata.
- VM contracts could support automated management, cloud interoperability, enterprise data-center migration, and bounded resource settings.
- SaaS integration research separates requirements into user-interface, process, and data integration to combine multiple SaaS products with in-house applications.
- A proposed Virtual Resource Mediation Layer abstracts CPU, memory, bandwidth, and storage access across public and private clouds, but vendor resistance makes implementation difficult.
- Most intercloud communication and resource-sharing research remains at the research-question stage without initial empirical results.
7. BUILDING CLOUDS
Research on building clouds covers management software, hardware provisioning, simulation, resilience, and energy-aware change management. The reviewed projects are diverse, and all identified building technologies address IaaS-level clouds rather than other service levels.
- Cloud construction research requires management software, hardware provisioning, simulators, and evaluation of management choices.
- OpenNebula provides unified virtual-resource management, VM lifecycle control, and configurable allocation policies, while Haizea manages resource leases.
- Containerized data-center building blocks use overprovisioning or cold nodes to tolerate component failures until performance falls below a replacement threshold.
- Simulation tools model data-center failures, middleware scalability, application performance overhead, scheduling, and virtualization effects.
- Energy-aware change-management work derives load from current SLAs, but cloud elasticity makes SLA-based load estimation difficult.
- The reviewed building-cloud papers pursue diverse research directions and all focus on IaaS-level clouds, with no identified technologies for other levels.
COMPUTING
Cloud computing enables new use cases by supplying elastic external resources for smartphones, large-scale testing, SaaS regression testing, and joint management of operating systems with applications.
- Cloud resources can overcome smartphone hardware limitations by partially offloading computation to external infrastructure.
- Cloud elasticity makes large-scale non-functional requirements testing feasible and affordable, though it does not fully replace testing against real managed endpoints.
- SaaS regression testing is necessary because each update becomes effective for all users, making exhaustive state testing impractical.
- IaaS clouds manage entire virtual appliances, deploying and maintaining operating systems jointly with their applications.
9. CONCLUSION
The reviewed academic work primarily advances cloud technology through standards, interoperability, and cloud design, while identifying unresolved technical obstacles and diverse research directions.
- Academic research has focused substantially on standards, interoperability, and methods for designing and building clouds.
- The review covers cloud definitions and socio-technical issues including costs, security, legal and privacy implications, and effects on IT work.
- Interoperability research targets machine-image, contextualization, and API-level compatibility across IaaS providers.
- Cloud growth faces obstacles including service and data lock-in, limited scalable storage, performance unpredictability, and data-transfer bottlenecks.
- The clearest research agenda concerns cloud interoperability, while building-cloud and cloud-use-case efforts remain diverse.