STAR Statement of Work for the Particle Physics Data Grid Collaboratory Pilot Extension
Jerome LAURET
Brookhaven National Laboratory
Upton, NY 11973-5000
To effectively move our users to the use of a robust Grid the STAR Collaboration plans to pursue its work on resource monitoring, Grid job submission, file replication and file placement but also will undergo several phases of software and infrastructure consolidation and GRID-ification.
We firmly believe that the success of job submission on the Grid depends on a common approach on describing the intent of a user rather than the way the job (or task) will be executed. While complex work flow and use of GRAM may be a way to describe the task at a fundamental level, the scientific community (and ultimately the end users) must have a way to either (1) transparently have access to Grid resources, (2) access those resources via a level of abstraction (a high level user description language U-JDL). Moreover, we believe that such a U-JDL should have the ability to interface with other services to make full use of resource brokering capabilities (resource discovery, computing on demand), advance intelligence in file placement and file movement, automated registration and interaction with Catalogs. For our community, the ability of tracing, controlling and in general modifying the task status at every point in time is also an important aspect of a job submission service (view of partial results from the output for example) and we plan to make use and integrate work and components addressing those issues such as troubleshooting and error recovery.
To that extent, we plan to deliver (in collaboration with other groups) a version of a U-JDL which will include the above considerations. We plan to develop and implement a prototype job submission service whose aim will be to serve as the core of Grid job submission. We believe the success of such a service to be tied to the success of Web Services either in its OGSA or WSRF form and plan to work with a general Web-Service approach with a later conformism with the solidified specifications. We will continue to work on fabric and queue (in the CS sense) monitoring which will serve as the basic information for our resource brokering algorithm. Later refinements may include service connected accounting information as available. We will seamlessly make use of existing work to provide our users with an analysis framework allowing for transparent use of Grid services. Integration work is foreseen as minimal in this area while design and requirement specification will be part of collaborative efforts.
A robust and reliable Grid can only be reliable to the extent it is traceable and secure with Grid users accountable. We view Authentication, Authorization and Accounting as fundamental components which will need to be addressed in a coherent fashion across the Grid. We plan to use components coming from other work in our Grid infrastructure and framework and will limit, to the extent it is achievable, ourselves to the evaluation, testing and implementation of such components.
Our current BNL/PPDG team is composed of Richard Casella, Eftratios Efstathiadis, Valeri Fine, Jerome Lauret and Dave Stampf. To meet our production and analysis challenges requiring hardening from the inside out, we plan to extend this team with a new hire.
Year-1
1st quarter: release of a first version of the requirement for a U-JDL, batch-based user analysis at some level and monitoring framework.
2nd quarter: support for SRM and Replica Registration Service operational testing, prototyping of a Job submission Service (based on U-JDL V1.x)
3rd quarter: Evaluation of Catalog and database access needed for data analysis, deployment of kernel toolkit to external sites. Network performance consolidation.
4th quarter: Consolidation of Grid job submission with SRM (file movement/placement) and testing of error recovery components.
Year-2
1st quarter: Interactive analysis framework deployment and testing
2nd quarter: Delivery of V2.x of the U-JDL specification including job tracking and control and interaction with more Grid components, prototyping of next version of the Job Submission Service.
3rd quarter: Accounting, Authorization component testing and integration.
4th quarter: Hardening of existing infrastructure.
Considering the magnitude of the task, PPDG funding will cover only part of the labor required to meet these goals; the remainder will come from STAR base program and BNL Information Technology Division (ITD) personnel. PPDG funds for each year will cover for
Task implementation (40% FTE, design and development)
Task implementation (15% FTE, testing, deployment, support)
Project Coordination (05% FTE)
Travel ($4000 for four trips year to PPDG meetings, at $1000 per trip – Berkeley)
No equipment is foreseen as covered by PPDG funding.
STAR has no unexpended funds from the original Particle Physics Data Grid project.