Application Skeletons: Encapsulating MTC Application Task Computation and I/O

Computer scientists who work on tools and systems meant to support or enable a variety of distributed computing applications want to prove that the systems they design actually help those applications. However, doing this by using the actual applications can be difficult due to policy or technical issues when accessing and building the application and necessary data sets. These issues led us to the idea of an Application Skeleton -- a simple yet powerful tool to build synthetic applications that represent real applications, with runtime, I/O, and intertask communication close to those of the real applications. This allows computer scientists to focus on the system they are building; they can work with the simpler Skeleton applications and be sure that their work will also be applicable to the real applications. Skeletons currently can create easy-to-access, easy-to-build, and easy-to-run bag-of-task, map-reduce, and multi-stage workflow applications. In this initial work, we show that a Skeleton version of the Montage application has a runtime difference of 2.6% in total on 64 processors on a BG/P supercomputer. And six of eight stages have an error within 5%.