PaperlayerView exact revision on arXiv

Source-linked AI summary

SLA-Oriented Resource Provisioning for Cloud Computing: Challenges, Architecture, and Solutions

Rajkumar Buyya, Saurabh Kumar Garg, Rodrigo N. Calheiros

arXiv:1201.4522v1cs.DCcs.NI

TL;DR

Existing data-center resource management does not adequately support SLA-oriented allocation or collectively address customer management, computational risk, and autonomic management. The paper proposes a market-oriented architecture implemented with Aneka and virtualization technologies for flexible, SLA-based provisioning. Prototype results show feasibility and effectiveness, while the authors call for deeper investigation of integrated management strategies.

  • Problem

    Existing data-center resource management systems do not support SLA-oriented allocation, and these management concerns have not been collectively incorporated into a market-based system.

  • Method

    The paper proposes a market-oriented SLA resource-management architecture integrating customer-driven service management, computational risk management, autonomic management, virtualization, and Aneka services.

  • Results

    The working Aneka prototype dynamically allocates resources to meet application QoS requirements while optimizing cost by allocating the minimum resources needed to meet deadlines.

  • Takeaways & Limitations

    SLA-oriented allocation provides a critical link for Cloud computing and can be implemented effectively using the Aneka platform.

  • Takeaways & Limitations

    The paper identifies a need for deeper investigation of strategies combining customer-driven service, computational risk, and autonomic Cloud management.

Abstract

from arXiv · show

Cloud computing systems promise to offer subscription-oriented, enterprise-quality computing services to users worldwide. With the increased demand for delivering services to a large number of users, they need to offer differentiated services to users and meet their quality expectations. Existing resource management systems in data centers are yet to support Service Level Agreement (SLA)-oriented resource allocation, and thus need to be enhanced to realize cloud computing and utility computing. In addition, no work has been done to collectively incorporate customer-driven service management, computational risk management, and autonomic resource management into a market-based resource management system to target the rapidly changing enterprise requirements of Cloud computing. This paper presents vision, challenges, and architectural elements of SLA-oriented resource management. The proposed architecture supports integration of marketbased provisioning policies and virtualisation technologies for flexible allocation of resources to applications. The performance results obtained from our working prototype system shows the feasibility and effectiveness of SLA-based resource provisioning in Clouds.

I. INTRODUCTION

Cloud and utility computing shift software delivery toward globally available, subscription-oriented services. This model reduces users’ infrastructure burden while requiring distributed data centers and networked virtual machines to provide reliable, responsive access.

  • I. INTRODUCTION: Utility computing offers computing services when needed, allowing users to pay for usage without maintaining dedicated infrastructure.Users can outsource jobs to computing service providers instead of investing heavily in their own infrastructure.
  • I. INTRODUCTION: Cloud computing aims to keep infrastructure and services available worldwide so companies can access business applications whenever needed.The model uses computing servers distributed across continents.
  • I. INTRODUCTION: Cloud computing can create data centers dynamically by assembling services of networked virtual machines.
  • I. INTRODUCTION: Providers distribute data centers geographically to provide backup during site failures and faster response through simultaneous workload distribution.

II. CHALLENGES AND REQUIREMENTS

SLA-oriented resource allocation is complex because providers must differentiate and satisfy many users’ service requests. Key requirements include customer-centered service quality and systematic analysis of computational risks.

  • II. CHALLENGES AND REQUIREMENTS: SLA-based management must differentiate and satisfy service requests according to users’ desired utility and service-quality parameters.SLAs also provide feedback mechanisms that can encourage or discourage service-request submissions.
  • II. CHALLENGES AND REQUIREMENTS: Customer satisfaction depends on service-quality factors such as communication, accessibility, personalized attention, and understanding specific customer needs.
  • II. CHALLENGES AND REQUIREMENTS: Data centers require broader study of customer characteristics to determine which characteristics should inform SLA-oriented resource allocation.
  • B. Computational Risk Management: Computational risk management must establish context, identify and assess risks, select management techniques, and create, implement, and review a risk plan.One example is the penalty and dissatisfaction risk from violating one customer’s SLA to fulfill another customer’s request.

C. Autonomic Resource Management

Changing service requirements require data centers to manage reservations, schedules, and prices continuously. Autonomic management and virtualization support dynamic adaptation to demand and differentiated user needs.

  • C. Autonomic Resource Management: Data centers must continuously monitor current requests, amend future requests, and adjust schedules and prices for new or changed requirements.
  • C. Autonomic Resource Management: Autonomic systems should self-configure components to satisfy new service requirements while managing limited resources, demand, and existing obligations.
  • D. SLA-oriented Resource Allocation Through Virtualization: Virtualization lets one physical machine host isolated logical VMs with different resource allocations, such as 10% and 20% of processing power.VMs can be started and stopped dynamically and assigned policies for different user needs.

E. Service Benchmarking and Measurement

Growing competition among Cloud providers increases the need for standardized service measurement. Benchmarks should represent realistic application and service requirements while supporting prediction of future user needs.

  • E. Service Benchmarking and Measurement: Cloud providers offer different computing services, making service measurement standards important for identifying services that satisfy customer needs.The Cloud Service Measurement Index Consortium has identified Service Measurement Indexes for evaluating Cloud services.
  • E. Service Benchmarking and Measurement: Resource-management policies require standardized benchmarks that reflect realistic application and service requirements.Such benchmarks can facilitate forecasting and prediction of future users’ needs.

F. System Modeling and Repeatable Evaluation

The paper uses simulation to evaluate SLA-oriented resource management strategies because real-world evaluation is difficult to make repeatable and controllable. CloudSim provides modeling capabilities for virtual resources, networks, application composition, discovery, and task execution.

  • Real-world evaluation is difficult because resources are distributed and service requests arrive from different customers at unpredictable times.
  • Discrete-event simulation is used to evaluate resource management strategies under varied operating scenarios.
  • CloudSim models Cloud resources, application scheduling, virtual resources, network connectivity, and resource configurations.
  • CloudSim also supports application composition, resource discovery, task assignment, and execution management for constructing evaluation models.

III. SLA-ORIENTED CLOUD COMPUTING VISION

The paper envisions SLA-oriented Cloud resource allocation centered on customer QoS, market mechanisms, computational risk management, autonomic management, and virtualization. Its proposed direction includes developing and implementing these models and policies in operational data centers.

  • Future work should develop SLA-oriented resource allocation models and policies designed specifically for data centers.
  • Market-oriented provisioning should allocate resources according to user QoS targets and workload demand patterns.
  • The vision combines customer-driven service management, computational risk management, market-based resource management, and autonomic management of changing requirements.
  • Virtual machines should dynamically assign resource shares according to service requirements.
  • The developed strategies and models are intended for implementation on real computing servers in operational data centers.

IV. STATE-OF-THE-ART

Existing resource-management and virtualization platforms provide important infrastructure capabilities, but the paper proposes an Aneka-based system specifically oriented toward SLA management, pricing, accounting, provisioning, and QoS.

  • Traditional systems such as Condor, LoadLeveler, LSF, and PBS use system-centric allocation focused on cluster performance and utilization.
  • Cloud platforms including Eucalyptus and OpenStack emerged to manage virtual machines and Cloud infrastructure.
  • OpenNebula provides dynamic allocation, advance provisioning, and scheduling through its Haizea lease scheduler.
  • Market-based resource management is intended to regulate supply and demand while providing economic incentives to users and providers.
  • The proposed Aneka-based system supports application scheduling, resource provisioning and monitoring, pricing, accounting, and QoS/SLA services in private and public Clouds.

V. SYSTEM ARCHITECTURE

The architecture places an SLA Resource Allocator between users or brokers and Cloud infrastructure, coordinating admission, autonomic resource management, pricing, accounting, SLA management, monitoring, dispatch, and virtual machines. These components support dynamic resource allocation while tracking service execution and fulfillment.

  • Users or brokers submit service requests to Cloud management systems for processing.
  • The high-level architecture contains an SLA Resource Allocator that interfaces between Cloud infrastructure and external users or brokers.
  • Admission control interprets QoS requirements and accepts or rejects requests to reduce overload and SLA violations.
  • Autonomic resource management uses VM migration and consolidation to adjust resources as application demands change.
  • Pricing manages demand, supports prioritization, and can vary by submission time, pricing rate, or resource availability.
  • Accounting records actual resource usage for charging, while SLA management tracks agreements and fulfillment history.
  • VM and application monitors track resource availability, entitlements, performance, and possible SLA breaches.
  • The Dispatcher deploys applications on appropriate virtual resources and initiates VM images on selected physical hosts.

VI. SLA PROVISIONING IN ANEKA

Aneka implements SLA-oriented provisioning through decoupled services that monitor jobs, estimate deadlines, and dynamically add or release resources. Its job-state workflow coordinates scheduling, resource pools, and feasibility decisions while accounting for resource-source constraints.

  • Architecture: Aneka is a Platform as a Service framework whose containers host services for scheduling, provisioning, accounting, and other management functions across heterogeneous resource pools.Its management layer handles clusters, public and private Clouds, and Desktop Grids.
  • Architecture: Decoupled Aneka services communicate through messages, allowing services to remain operational when another service is disabled.The IService and ServiceBase abstractions provide common lifecycle, logging, and event-handling behavior.
  • SLA management: The Scheduler and Provisioning services enable SLA-driven execution, while admission control interprets QoS requirements and checks resource and workload status before accepting requests.SLAs are expressed as application deadlines supplemented by user-provided task-runtime estimates.
  • Dynamic provisioning: Aneka updates runtime estimates and provisions additional resources when current capacity cannot meet a job deadline, or releases excess resources when fewer suffice.The provisioning decision responds to completed tasks and newly received jobs.
  • Job lifecycle: Jobs can move among QoS, queued, provisioned, underprovisioned, unfeasible, and finished states as resource availability and deadline feasibility change.An unfeasible job receives no additional allocations, and deadlines smaller than resource boot time can make jobs unfeasible.
  • Resource lifecycle: Resource decommissioning is delegated to Resource Pools because procedures differ by resource source, while public-Cloud resources may remain active until the paid billing window ends.This coordination supports execution within user-defined deadlines and resource reuse across jobs.

VII. PERFORMANCE EVALUATION

The evaluation tests Aneka’s SLA provisioning on Amazon EC2 using a CPU-intensive workload and deadlines of 45, 30, and 15 minutes. Results indicate that dynamic allocation meets QoS requirements while minimizing provisioned resources, although more aggressive time optimization could reduce execution time at higher cost.

  • Experimental setup: The evaluation was conducted entirely on Amazon EC2 in the USA East Coast using static and dynamic resources.The static setup included one m1.large Aneka master and four m1.small Aneka workers.
  • Workload: The CPU-intensive workload used 120 tasks per job, with each task assigned an execution time of 2 minutes.A job would therefore require 4 hours on a single machine.
  • Workload: Aneka was tested without QoS and with user-defined deadlines of 45, 30, and 15 minutes.These configurations were compared in Table 1.
  • Results: Dynamic resource allocation effectively met application QoS requirements while allocating the minimum resources needed to meet each deadline.Execution times were reported as very close to the specified deadlines, supporting the paper’s cost-optimization strategy.
  • Results: More aggressive time-based optimization could meet deadlines by larger margins, but it would incur greater user cost and is left for future research.The current strategy prioritizes cost optimization over minimizing execution time.

VIII. CONCLUSIONS AND FUTURE DIRECTIONS

The paper frames SLA-oriented allocation as a key challenge for complex, fast-changing Cloud applications and presents Aneka as an implementation of the proposed framework. It concludes that deeper integration of customer-driven management, computational risk management, and autonomic Cloud management remains necessary.

  • Challenges: SLA-oriented allocation must address complex applications that require multiple services and collaboration among organizations or businesses.Such applications depend on specific services to function successfully.
  • Significance: Fast turnaround requirements make SLA-oriented resource allocation a critical link to the success of the next ICT era of Cloud computing.The paper connects this need to increasingly competitive business environments.
  • Implementation: The proposed framework was shown to be implementable using the Aneka platform.The conclusion presents Aneka as the practical realization of the framework described in the paper.
  • Future directions: Future work should investigate customer-driven service management, computational risk management, and autonomic Cloud management within SLA-oriented allocation strategies.The stated goals are improved system efficiency, fewer SLA violations, and better service-provider profitability.
Loading 1201.4522v1…