Enabling the Co-Allocation of Grid Data Transfers

  • Authors:
  • Sudharshan Vazhkudai

  • Affiliations:
  • -

  • Venue:
  • GRID '03 Proceedings of the 4th International Workshop on Grid Computing
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

Data-sharing scientific communities use storage systems asdistributed data stores by replicating content. In such highlyreplicated environments, a particular dataset can reside atmultiple locations and can thus be downloaded from any oneof them. Since datasets of interest are significantly large insize, improving download speeds either by server selection orby co-allocation can offer substantial benefits.In this paper, we present an architecture for co-allocatingGrid data transfers across multiple connections, enabling theparallel download of datasets from multiple servers. We havedeveloped several co-allocation strategies comprising ofsimple brute-force, history-based and dynamic load balancingtechniques as a means both to exploit rate differences amongthe various client-server links and to address dynamic ratefluctuations. We evaluate our approaches using the GridFTPdata movement protocol in a wide-area testbed and presentour results.