Dr. David Buján Carballal

An open problem within the domain of service oriented architectures, such as grid infrastructures, is the discovery of resources and/or services suitable for the execution of a specific task. Grid information producers generally provide functional information on the resources they monitor. Therefore information models represent this syntactic information whereas information consumers use such functional properties in order to select resources. In practice many jobs are generally restarted due to errors in the allocated resources. However various initiatives which try to use ad-hoc techniques in order to manipulate non-functional properties have been proposed. This thesis proposes a new approach in the modelling of non-functional properties of grid resources, so that non-functional properties can be used both by the producers and the consumers of grid-based information in order to improve both the performance of the infrastructure layer and the performance of the jobs executed on such infrastructure. Various innovations have been required regarding this new approach. In the first place, a model based on semantic technologies has been developed in order to integrate both the existing models focused on representation of grid-based information and those focused on non-functional properties, with the aim of enriching the existing information on every resource at every level present on each grid node. Secondly, a new methodology for the representation of non-functional properties in the grid domain has been proposed so that it can be used by the administrators of the various nodes in order to enrich the definition of their resources and improve the user experience with respect to the execution of their jobs. Thirdly, such methodology has been applied to the representation of three non-functional properties: two at computational resource level (availability and reliability) and one at grid node level (reliability). Finally, a set of algorithms modified by including non-functional properties at computational resource level which allows the discovery and selection of such resources within the context of a testing model for a grid simulation tool has been developed. This set of modified algorithms allows the comparison of the results obtained by using such non-functional properties in a stand-alone or combined fashion during the discovery or selection of resources.