A Component Architecture for the Message Passing Interface (MPI): The Systems Services Interface (SSI) for LAM/MPI

This work presents the design and implementation of a component system architecture in LAM/MPI, a production quality, open source implementation of the MPI-1 and MPI-2 standards. Previous versions of LAM/MPI, as well as other MPI implementations, are based on monolithic software architectures that -- regardless of how well-abstracted and logically constructed -- are highly complex software packages, presenting a steep learning curve for new developers and third parties. As a result, parallel researchers face enormous logistical and technical difficulties when using or adapting existing implementations for their own work. Not only are existing code bases typically locked into highly-specific implementation models (effectively preventing extensions that did not already conform to existing models), the time investment required to train a researcher in a complex software system can be prohibitive. To address these issues, the current version of LAM/MPI has been re-architected to utilize a component system architecture consisting of four component frameworks and a meta framework that ties them together. Each component framework was designed from analysis of prior monolithic implementations of LAM/MPI and represents a major functional category: run-time environment startup, MPI point-to-point communication, MPI collective communication, and parallel checkpoint/restart. The result is an MPI implementation this is highly modular, has published abstraction and interface boundaries, and is significantly easier to develop, maintain, and use as a vehicle for research. Performance results are shown demonstrating that this component-based approach provides identical (if not better) performance compared to prior monolithic-based implementations.