Exploiting Dynamic Deployment in a Distributed Query Processor for the Grid

  1. Lookup NU author(s)
  2. Dr Arijit Mukherjee
  3. Professor Paul Watson
Author(s)Mukherjee A, Watson P
Publication type Report
Series TitleSchool of Computing Science Technical Report Series
Year2008
DateFebruary 2008
Report Number1068
Pages11
Full text is available for this publication:
Grid computing has enabled users to perform computationally expensive applications on distributed resources acquired dynamically. It has also allowed users to combine data and analysis components into new applications from sites all over the world. Often such distributed data is structured, and an established way of structuring computations is found in distributed query processing. The established Grid tools like OGSA-DAI and OGSA-DQP provide respectively a common interface to different underlying databases and a way of expressing a global schema which unites the distributed resources and supports evaluation of declarative queries in distributed fashion. While Grid computing offers these significant benefits to a user, it also incurs the difficulties that arise through distribution, especially at wide area scale, on non-dedicated resources. Computation and communication costs can both be significant and vary significantly. In such contexts, a range of techniques for dynamic adaptation of distributed queries need to be employed in order to ensure reasonably efficient execution. DynaSOAr is an infrastructure for dynamically deploying web services over a set of networked resources with virtualization features thereby creating support for ad-hoc Virtual Organisations. This paper investigates the possibilities of exploiting dynamic service deployment techniques provided by DynaSOAr in the established OGSA-DQP system and describes a distributed query processing framework with support for dynamic deployment which allows the demand-driven deployment of the query processing engine, analysis services and databases deployed within virtual machines on an internet scale. The results of the internet-scale experiments underline the effectiveness of such a framework in improving the performance by introducing parallelism, reducing communication costs and dealing with sudden unavailability of resources.
InstitutionSchool of Computing Science, University of Newcastle upon Tyne
Place PublishedNewcastle upon Tyne
URLhttp://www.cs.ncl.ac.uk/publications/trs/papers/1068.pdf
ActionsLink to this publication