Distributed computing - grid scheduling techniques

Definition : Application-aware grid scheduling

(c) www.csc.fi/grid

Application-aware scheduling algorithms can be used for efficiently distributing and managing parameterized computer experiments on multi-core processors, computer clusters and grids. The number of (computational expensive) simulations is automatically reduced.

Background

During the design and analysis of complex systems in science and engineering, sophisticated and realistic numerical simulations are commonly used to limit the number of expensive prototypes and real-life measurements. Computer simulation provides a means to mimic the behaviors of complex real systems both quickly and economically.

High performance computer grids and clusters provide cost-effective problem-solving power, and are widely used by scientists and engineers to simulate the behavior of complex systems as a function of multiple parameters, and to build scalable models. [ref]

Research goals

We study and develop reliable adaptive scheduling techniques for efficient parametric simulation of the dynamics of large and complex systems on multi-core processors, computer clusters and grids, based on automatically gathered application-specific information. No prior knowledge of the application is required.

This research is based on, or linked with:
  • Distributed/parallel computing
    [Scheduling and Load Balancing, Performance Evaluation and Modeling, Middleware, High Performance Computing (HPC)]

  • Computer-aided Design
    [Design and Analysis of Computer Experiments (DACE), Metamodeling , Data-driven information processing]

  • Artificial Intelligence (AI)
    [Machine Learning (ML), Adaptive/Sequential sampling, Reflective Exploration (RE)]