A monitoring service for large-scale dynamic query optimisation in a grid environment

  • Authors:

  • Mahmoud El Samad;Julien Gossa;Franck Morvan;Abdelkader Hameurlain;Jean-Marc Pierson;Lionel Brunie

  • Affiliations:

  • Institute of Research in Computer Science of Toulouse (IRIT), Paul Sabatier University (UPS), Toulouse, UMR5505, France.;Laboratory of Computer Science in Image and Information Systems (LIRIS), National Institute of Applied Science (INSA) of Lyon, UMR5205, France.;Institute of Research in Computer Science of Toulouse (IRIT), Paul Sabatier University (UPS), Toulouse, UMR5505, France.;Institute of Research in Computer Science of Toulouse (IRIT), Paul Sabatier University (UPS), Toulouse, UMR5505, France.;Institute of Research in Computer Science of Toulouse (IRIT), Paul Sabatier University (UPS), Toulouse, UMR5505, France.;Laboratory of Computer Science in Image and Information Systems (LIRIS), National Institute of Applied Science (INSA) of Lyon, UMR5205, France

  • Venue:
  • International Journal of Web and Grid Services
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

The execution plans generated by the traditional optimisers for large-scale distributed queries in a grid can be suboptimal for the following reasons: the centralisation of decisions made by the optimiser; the inaccuracy of estimations; the unavailability of up-to-date description of resources. In this paper, we propose an approach to improve the estimation of the execution cost of a query (or part of a query) in a grid environment by using Mobile Agents (MAs) and runtime monitoring information. First, Mas allow dynamic optimisation in a decentralised and autonomous way. Second, the retrieval of the monitoring information causes some issues addressed by a dedicated service called the Network Distance Service (NDS). The performance evaluation shows that our approach allows a better estimation of the execution cost of a query in a grid and a large benefit to monitor the CPU for a mobile join when the subestimation errors increase while the benefit to monitor the bandwidth is poor.