Pi: A Parallel Architecture Interface for Multi-Model Execution
Abstract
Existing parallel architectures are constructed monolithically, with no well defined boundaries separating model and machine issues. This makes it difficult to evaluate the effect of a single component of an architecture, or compare it with the corresponding components of other architectures. Machine hardware is specialized to support a single programming model, even though similar but more general mechanisms could support a variety of models. Often details of an implementation become visible in the programming environment, restricting future implementation improvements because of compatibility. The lack of an interface between model and machine issues also complicates the translation of a machine improvement into a performance improvement for model applications. This thesis defines Pi, a parallel architecture interface that separates model and machine issues, allowing them to be addressed independently. This provides greater flexibility for both the model and machine builder. Pi addresses a set of common parallel model requirements including low latency communication, fast task switching, low cost synchronization, efficient storage management, the ability to exploit locality, and efficient support for sequential code. Since Pi provides generic parallel operations, it can efficiently support for sequential code. Since Pi provides generic parallel operations, it can efficiently support many parallel programming models including hybrids of existing models. Pi also forms a basis of comparison for architectural components. (kr)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1990
- Accession Number
- ADA228345
Entities
People
- Donald S. Wills
Organizations
- Massachusetts Institute of Technology