Overview

• We developed three new competitive job assignment solutions for metacomputers.

• We demonstrated that these strategies improve over standard strategies using two performance measures: “average slowdown” and ability to meet real time demands.

• We developed two new metacomputing platforms using these strategies.

Overview

Previous slide

Next slide

Back to first slide

View graphic version

Our Three New Metacomputing Strategies Were:

• The Enhanced PVM Strategy, assuming that job assignment is permanent and a job’s memory requirement is known at its arrival;

• The Differential PVM Strategy, assuming that job assignment is permanent and a job’s memory requirement is not known at its arrival; and

• The Enhanced Mosix Strategy, assuming that the system can reassign jobs after their arrival and memory requirements are known or quickly discovered.

Our Experiments showed that these strategies perform very well in terms of average resource usage.

The Theory demonstrates that these strategies are competitive with the optimal algorithm in maximum resource usage.

Implementation occurred on two platforms:

• The Java Market for Java-based metacomputing on the Internet; and

• The Frugal System for Jini-based local and global metacomputing.