A hierarchical and multiscale approach to analyze E-business workloads

  • Authors:

  • D. A. Menascé;V. A. F. Almeida;R. Riedi;F. Ribeiro;R. Fonseca;W. Meira, Jr.

  • Affiliations:

  • Department of Computer Science, MS 4A5, George Mason University, 4400 University Drive, Fairfax, VA;Department of Computer Science, Federal University of Minas Gerais, Belo Horizonte, MG 31270, Brazil;Department of Electrical and Computer Engineering, Rice University, Houston, TX;Department of Computer Science, Federal University of Minas Gerais, Belo Horizonte, MG 31270, Brazil;Department of Computer Science, Federal University of Minas Gerais, Belo Horizonte, MG 31270, Brazil;Department of Computer Science, Federal University of Minas Gerais, Belo Horizonte, MG 31270, Brazil

  • Venue:
  • Performance Evaluation
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

Understanding the characteristics of electronic business (E-business) workloads is a crucial step to improve the quality of service offered to customers in E-business environments. This paper proposes a hierarchical and multiple time scale approach to characterize E-business workloads. The three levels of the hierarchy are user, application, and protocol, and are associated with customer sessions, functions requested, and HTTP requests, respectively. Within each layer, an analysis across several time scales is conducted. The approach is illustrated by presenting a detailed characterization of two actual E-business sites: an online bookstore and an electronic auction site. Our analysis of the workloads showed that the session length, measured in number of requests to execute E-business functions, is heavy-tailed, especially for sites subject to requests generated by robots. An overwhelming majority of the sessions consist of only a handful requests, which seems to suggest that most customers are human (as opposed to robots). A significant fraction of the functions requested by customers were found to be product selection functions as opposed to product ordering. An analysis of the popularity of search terms revealed that it follows a Zipf distribution. However, Zipf's law as applied to E-business is time scale dependent due to the shift in popularity of search terms. We also found that requests to execute frequent E-business functions exhibit a pattern similar to the HTTP request arrival process. Finally, we demonstrated that there is a strong correlation in the arrival process at the HTTP request level. These correlations are particularly stronger at intermediate time scales of a few minutes.