Source-linked AI summary
Cloud Computing and Grid Computing 360-Degree Compared
Ian Foster, Yong Zhao, Ioan Raicu, Shiyong Lu
TL;DR
Cloud Computing lacks a consensus definition and remains closely related to Grid Computing. This paper compares the two across multiple dimensions, finding substantial commonality alongside important differences and identifying challenges and opportunities for their evolution.
Problem
Cloud Computing lacks a consensus definition, while its relationship with Grid Computing and related paradigms requires comparison across essential characteristics.
Method
The paper compares Cloud and Grid Computing across vision, architecture, technology, security, programming, business, compute, data, application, and abstraction models.
Results
Clouds and Grids share substantial commonality but differ across security, programming, business, compute, data, application, and abstraction models.
Takeaways & Limitations
Close comparison can help both communities understand, share, and evolve infrastructure and technology, accelerating Cloud Computing from early prototypes toward production systems.
Takeaways & Limitations
Cloud security is a major adoption concern, including risks involving privileged access, regulatory compliance, data location, and data segregation.
Abstract
from arXiv · showhide
Cloud Computing has become another buzzword after Web 2.0. However, there are dozens of different definitions for Cloud Computing and there seems to be no consensus on what a Cloud is. On the other hand, Cloud Computing is not a completely new concept; it has intricate connection to the relatively new but thirteen-year established Grid Computing paradigm, and other relevant technologies such as utility computing, cluster computing, and distributed systems in general. This paper strives to compare and contrast Cloud Computing with Grid Computing from various angles and give insights into the essential characteristics of both.
1.1 Defining Cloud Computing
Cloud Computing lacks consensus, so the paper defines it as a large-scale, Internet-delivered distributed paradigm built around abstracted, virtualized, dynamically scalable, managed resources. It overlaps with and evolved from Grid Computing, but today’s Clouds do not clearly satisfy all of Grid’s defining criteria.
- Cloud Computing definition: Cloud Computing is a large-scale distributed paradigm delivering abstracted, virtualized, dynamically scalable, managed computing resources and services on demand over the Internet.The definition is driven by economies of scale and serves external customers.
- Motivations for Cloud adoption: Cloud adoption is driven by falling hardware costs, rising computing and storage capacity, exponentially growing data, and widespread Services Computing and Web 2.0 use.These pressures arise from increasing computing and storage needs in the Internet Age.
- Relationship to Grid Computing: Cloud Computing overlaps with Grid, Utility, Services, and general distributed computing, while the paper argues that it evolved from and relies on Grid infrastructure.This relationship reflects a shift in focus from infrastructure delivery toward services.
- Grid Computing definition: Grid Computing enables federated resource sharing and coordinated problem solving across dynamic, multi-institutional virtual organizations.Its infrastructure spans physically distributed institutions or logically related projects and groups.
- Cloud–Grid distinction: Foster’s Grid checklist requires decentralized resource coordination, open general-purpose protocols and interfaces, and non-trivial quality of service.Although quality of service applies to Cloud Computing, decentralization and open protocols are unclear for today’s Clouds.
2 Comparing Grids and Clouds Side-by-Side
The section compares Clouds and Grids across architecture, programming models, data management, applications, infrastructure, security, and provenance. Clouds offer layered services and centralized resources, while Grids support heterogeneous, multi-domain scientific workloads but face interoperability, monitoring, and security challenges.
- Architecture and services: Cloud Computing is organized into four layers: fabric, unified resource, platform, and application.Clouds also expose IaaS, PaaS, and SaaS, but interfaces between these service levels remain insufficiently standardized.
- Data management: Data locality and joint compute-data management are essential for scalable Cloud and Grid applications because repeated data movement becomes a bottleneck.Distributing data across computers and steering computation toward it minimizes movement and improves performance and scalability.
- Programming models: Grid programming must handle multiple administrative domains, heterogeneous and dynamic resources, exceptions, and large-scale scientific computation.Workflow systems suit loosely coupled computations centered on managing many datasets and tasks, while MapReduce provides map and reduce functions for large datasets.
- Applications: Clouds can support many applications, but HPC workloads requiring fast, low-latency interconnects remain better suited to Grids.Cloud applications are expected to be largely loosely coupled and transactional, although their application landscape remains immature.
- Infrastructure and interoperability: Clouds generally use more homogeneous, organization-controlled data centers, whereas Grids span heterogeneous administrative domains, making cross-provider interoperability difficult.Cloud interfaces and platform technologies can also be proprietary, complicating integration across services and providers.
- Security, monitoring, and provenance: Clouds face comparatively weaker security and less straightforward monitoring than Grids, while provenance requires open, universal representations and query models.Cloud security concerns include privileged access and regulatory compliance; Grid trust models allow identity-delegated access across sites.
3 Conclusions and lights to the future
Clouds and Grids share substantial vision, architecture, and technology, but differ across security, models, applications, and abstractions. The future calls for interoperable, on-demand distributed infrastructure combining Cloud-scale utilities with Grid distribution, while current commercial clouds remain proprietary and specialized.
- Conclusions: Clouds and Grids share vision, architecture, and technology but differ in security, programming, business, compute, and data models, applications, and abstractions.The comparison also identifies challenges and opportunities intended to help both communities understand, share, and evolve infrastructure and technology.
- Future directions: Future computing grids are expected to combine small-scale producers, such as local clusters and embedded processors, with large utilities.This prediction is based on computing economics becoming increasingly similar to energy economics.
- Future directions: Distributed Clouds or Grids will need on-demand provisioning and configuration of integrated virtual systems tailored to end-user capabilities.The required systems must provide precise capabilities needed by an end-user.
- Future directions: Future protocols and tools must support provider discovery, demand handoff, reservation management, payment, resource management, and distributed computation.The infrastructure must combine the centralized scale of Cloud utilities with the distribution and interoperability of Grid facilities.
- Current limitations: Commercial-cloud methods remain mostly proprietary and specialized, while Grid technologies and protocols address interoperability between providers.The paper notes that provider interoperability has not yet emerged as a goal in commercial clouds, whose methods often serve internal uses such as large-scale data analysis.
- Future directions: Industry, academia, government laboratories, and future communities may all contribute protocols and tools, but their collaboration remains uncertain.These communities may either find common cause or proceed along parallel paths.