Computing on the Net
    
                  Francine Berman

  Networked "computers", often called "Grids", can be comprised of distributed 
computational sites, remote instruments, data archives, networks, and other 
resources.  Grids are becoming increasingly important and prevalent as a 
platform for computation.  However, application performance is often difficult
to achieve on Grids:  resources are heterogeneous, reside in distinct
administrative domains, and are shared by other applications which cause
the deliverable performance of these resources to change dynamically.
Achieving application performance in this environment is challenging, and
requires careful scheduling and adaptation to the dynamic performance
fluctuations of all relevant Grid resources.

  AppLeS (Application-Level Scheduler) is a project whose goal
is to develop and deploy custom performance-oriented application
schedulers for individual Grid and cluster applications, as well as
scheduling templates for common classes of Grid and cluster applications.
In this talk, we describe the AppLeS approach to developing application
schedulers.  We focus on current work in the project and discuss new
directions for the project over the next few years.