Abstract
Wide-area computational grids have the potential to provide large amounts of computing capacity to the scientific community. Realizing this potential requires intelligent data management, enabling applications to harness remote computing resources with minimal remote data access overhead. We define execution domains, a framework which defines an affinity between CPU and data resources in the grid, so applications are scheduled to run on CPUs which have the needed access to datasets and storage devices. The framework also includes domain managers, agents which dynamically adjust the execution domain configuration to support the efficient execution of grid applications. In this paper, we present the execution domain framework and show how we apply it in the Condor resource management system.
Original language | English (US) |
---|---|
Pages (from-to) | 246-252 |
Number of pages | 7 |
Journal | Computer Physics Communications |
Volume | 140 |
Issue number | 1-2 |
DOIs | |
State | Published - Oct 15 2001 |
Externally published | Yes |
Keywords
- Checkpointing
- Cluster computing
- Computational grids
- Condor
- High throughput computing
- Network scheduling
- Remote I/O
ASJC Scopus subject areas
- Hardware and Architecture
- General Physics and Astronomy