| Abstract | This thesis presents static code analysis approach to gathering information about
remote tasks to be executed by applications using GridRPC API for collective
mapping.
Collective mapping of tasks to servers has significant performance benefits
over mapping individual tasks independently. However, it requires information
about remote tasks an application is going to execute to be available before the
application starts executing these remote tasks.
This thesis presents static code analysis approach, which allows collecting essential
part of this information by analysing code during compilation. The application
is modified automatically by preprocessor to utilise collected information
and build optimal task-to-server mapping during runtime.
Advantages of static code analysis over other approaches to collective mapping
are outlined, and experimental results using real world application are presented. |
