|
|
MW
Software Framework for Master-Worker Metacomputing Applications.
Jean-Pierre Goux, Jeff Linderoth,
Sanjeev Kulkani and Mike Yoder
Hardware resources in a typical metacomputer are dynamically available
and heterogeneous. To parallelize existing algorithms on such
platforms, users must address issues such as fault tolerance, task
scheduling, and interprocess communication. By handling some of these
issues automatically and exposing others, MW provides an API for
rapidly implementing master-worker algorithms in metacomputing
environments. MW has been used to implement algorithms with complex
control structures such as branch-and-bound by mapping such structures
onto the master-worker paradigm with little loss of efficiency. MW
uses the opportunistic resource management and fault tolerance
facilities of PVM, to provide a powerful virtual platform. MW
is structured so that extensions to other resource management and
message passing protocols is straightforward. A file-based communication
version of MW has been recently released for applications with large
grainsize.
MW is implemented as a set of C++ abstract base classes. Three
abstract base classes must be reimplemented for each specific
application. The MWDriver class corresponds to the master process and
contains the control mechanisms for distributing tasks to workers.
The MWTask class describes the inputs and outputs-the data and
results-that are associated with a single unit of work. The
MWWorker class initializes each worker process and executes any tasks
that are sent to it by the master. The relationships between these
classes are depicted in the figure below.
More information about MW :
|
