This collaborative project between EPSRC (Engineering and Physical Sciences Research Council) and PPARC (Particle Physics and Astronomy Research Council) is funded through the UK e-Science programme. Its goal is to engage the UK computer science community with EU DataGrid related activities within the UK. Computer scientists and High Energy Physicists are working together within the London e-Science Centre (LeSC) to investigate how the existing service oriented architecture of the EU DataGrid could be migrated to the Open Grid Services Architecture (OGSA).

Job Description Markup Language

All jobs within the EU-Datagrid are specified in a Job Description Language (JDL) derived from the syntax used within Condor ClassAds. LeSC has been examining the JDL to see how this work could be utilised by the wider Grid community. This work is being undertaken as web services, which form the basis of OGSA, are beginning to be adopted as a Grid middleware. An initial result of this work has been the development of an XML version of the JDL called JDML, which is now being used to define jobs for execution on specific compute resources within the ICENI environment. A job is described in JDML and sent to an OGSA compatible service for mapping to a resource and subsequent execution. A draft document describing the JDML can be found here.

Web Service representation of EU DataGrid architecture

Load Balancing through OGSA

OGSA will provide an interoperable set of protocols, based on open industry supported frameworks (XML, SOAP, WSDL & UDDI) that can be used to deliver Grid services in a platform independent manner. We are developing a Job Submission Service as a mean to provide session-based utilisation of an underlying resource as a transient service. The virtualisation of computational resources as services introduces a layer where middleware can gather load information in order to make informed selections for multi-user multi-job resource utilisation.

At LeSC, we are experimenting with building load-balancing OGSA factory services that instantiate service instances by delegating the creation request to a pool of distributed factory instances through a defined load-balancing strategy. This generic framework separates the decision-maker from the information gatherer, therefore providing an extensible middleware approach for introducing intelligent resource virtualisation.

A user (1) submits a job described in JDML to a Virtual Job Submission Service. This request is delegated to one of many back end resources (2) for service instantiation (3). The Grid Service Handle is returned to the user (4) for further direct interaction with the service instance (5).

This infrastructure will be used to investigate load balancing and throughput issues within Work Package 1 of the EU-DataGrid project relating to job submission through the resource broker.