A novel approach for service performance analysis and forecast

  • Authors:
  • Sid Kargupta;Sue Black

  • Affiliations:
  • University College London;University College London

  • Venue:
  • Journal of Web Engineering
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

This research establishes a predictive model to forecast the impact on service performance for changes to the underlying activities of the service's components. It deduces a relational model between a service's performance, its application component latencies and the request load. The major challenge the IT industry is currently facing with the cost associated with repeated performance testing to modify live systems has been addressed. The notion of implicit Operation Impedance gradient (IG) and Operation Potential (V) in Service Provider-Consumer contracts has been introduced. This work establishes that 'IG', which impacts the overall Operation Performance (P), is influenced by the underlying application components' activities in distinct patterns. A high-level runtime abstract model is empirically deduced between 'IG', 'V' and 'P' by applying established mathematical techniques. Model based indicative values of some features are computed and associated with the actual empirical values of other features against various system configurations. Appropriate regression types are applied to enable trend extrapolation/interpolation. The datasets affirmed effectiveness of the model to assess impact of modifications to the underlying application components on the operation's performance without repetitive full scale external performance/benchmark testing. This also enables fine tuning of application components to retrofit prescribed Quality of Services. To address real life applications, this paper describes a Matrix based technique used for the assessment of changes to multiple types of application component activities simultaneously. The method of calibrating the Matrix aided model has also been discussed briefly.