The HEPGrid Project - High Energy Physics and Grid Network

Introduction

This project aims to design and develop a safe, secure, dynamic and adaptive Data Grid Resource Broker for managing and scheduling data intensive science applications on geographically distributed resources. The broker will engineered to logically couple with data sources to perform online extraction data from closest data sources or replicas depending on user requirements. The broker selects data sources depending on their cost and user defined quality of services to optimize the use of resources. It evaluates whether to process on a resource where data is available by moving the application code, moving data to a resource where the application is available, or move the data and the application to a resource whether computing power most economically is available. This is a significant challenge in the design and evaluation of scheduling algorithms. These will then be incorporated into application scheduler. The requirements for the Data Grid Broker scheduler will be derived from two application areas: High Energy Physics and Molecular Modeling for Drug Design.

Building upon the earlier work of Experimental Particle Physics Group at the the University of Melbourne, we plan to port the Belle Experiment analysis code to the GRID framework. Belle is an experiment based at the KEK, Laboratory in Japan that is right now accumulating tens of terabytes of data per year. Our plan is to initially test the suitability of the GRID framework to handle a real running experiment and then to be deployed as the computing solution for the Belle experiment.

The proposed Kangaroo Data Grid broker leverages existing grid middleware systems such as Globus and Nimrod-G. The framework for Problem Solving Environment builds on existing works and extends its capability to support tasks grouping, dependencies and flow graphs. The user-level data management and caching techniques will be developed for smart scheduling and also for sharing results of event processing with other researchers.

The Team Members

Publications

  • Under Construction!

Base Platform

  • Existing data analysis software developed by the Belle and ATLAS collaborations.
  • Physics Application Code: Belle: 800,000 lines of code (C++)and ATLAS: 1.5 million lines of code (C++)·
  • Virtual Lab: Tools developed to enable Drug Design on Grid to be used to handle some aspects of data management.
  • GridSim Toolkit: for design and evaluation of scheduling ideas!
  • Nimrod: for Exploring the use of Nimrod Parameter Specific Laguage (PSL) for creating parameter sweep application.
  • Nimrod-G Grid Resource Broker: for computational job scheduling.
  • Globus Toolkit: for having uniform access to diverse remote resources.
  • World Wide Grid (WWG) serves as testbed for research.