Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Resource and performance prediction at high utilization for n-tier cloud-based service systems
Zhang W., Shi Y., Zheng Y., Liu L., Cui L.  ACSW 2017 (Proceedings of the Australasian Computer Science Week Multiconference, Geelong, Australia, Jan 30-Feb 3, 2017)Article-No. 43.2017.Type:Proceedings
Date Reviewed: Apr 27 2017

Cloud provision providers offer pertinent services such as software-as-a-service (SaaS), platform-as-a-service (PaaS), and infrastructure-as-a-service (IaaS) that require users to sign service-level agreements (SLAs). Unless there are adequate tools for forecasting and meeting the requirements of the resources stipulated in various SLAs, cloud service providers will continue to be prone to litigation. But what are the characteristics of highly used cloud resources, and how should they be forecasted to minimize the violations of SLAs and litigation? Zhang et al. present quantitative models for estimating the response times and highly used resources in cloud computing systems.

The authors echoed the shortfall of the available prediction models in the literature to accurately estimate (1) the round-trip response times of resource utilization data in the cloud, (2) the response times to process complicated queries, and (3) the highly utilized resources in cloud environments. Consequently, they propose a resource and performance prediction model for investigating the mixtures of transaction workload traffic in web, application, and database servers.

The framework for resource and performance includes three models. A prototypical benchmark system is used to generate a mixture of workloads that mirrors several simultaneously executing heterogeneous queries in a cloud computing environment. The benchmark system produces a representative resource used to process queries and gathers performance data for modeling resource utilization and system performance. The familiar kernel canonical correlation analysis (KCCA) model is used to explore the associations between resource usage and workload. A Gaussian process with non-parametric models is used to recognize haphazard functions over dissimilar workloads and query mixtures to forecast response times.

The effectiveness of the resource and performance prediction models was investigated in a lifelike environment with servers that contain web, application, and data layers. In experiments, the benchmark request types were randomly assigned to servers for retrieval of query results. One thousand experimental data points on central processing unit (CPU), memory, disk, and cache and response times were used to gauge the accuracy of the Gaussian and KCCA models. The Gaussian model achieved a high average prediction of the actual average response time, particularly under huge workload situations. The KCCA model attained highly accurate resource usage predictions at various layers of a system, even though the model produced significantly higher forecast results for CPU and disk than for memory and cache usage. To what extent are cloud service providers satisfying the resource requirements in SLAs? Undoubtedly, the authors provide great insights into the kinds of data for investigating this nontrivial question. All cloud service providers and users should read this paper.

Reviewer:  Amos Olagunju Review #: CR145229 (1707-0470)
Bookmark and Share
  Reviewer Selected
Featured Reviewer
 
 
Metrics (D.2.8 )
 
Would you recommend this review?
yes
no
Other reviews under "Metrics": Date
A comparison of time domains for software reliability models
Musa J., Okumoto K. Journal of Systems and Software 4(4): 277-287, 1984. Type: Article
May 1 1985
On software equations
Král J. Information Processing Letters 19(4): 191-196, 1984. Type: Article
Jun 1 1985
Software metrics: establishing a company-wide program
Grady R., Caswell D., Prentice-Hall, Inc., Upper Saddle River, NJ, 1987. Type: Book (9789780138218447)
Apr 1 1988
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy