Resource Management for the Tagged Token Dataflow Architecture.
Abstract
The Taged Token Dataflow Architecture is multiprocessor based on the U-interpreter model of dataflow computation. It captures the essential execution mechhanism of the U-interpreter precisely; operations are enabled for execution by the availability of operand data. However, computational resources in the model and the machine are viewed quite differently. This thesis addresses four major resource management issues essential to bridge the gap between the U-interpreter and the Tagged Token Dataflow Architecture: Termination detection; Token store overflow; Iteration identifier overflow; and Program deadlock. This thesis offers a way to overcome the differences between the model and the machine through a concerted approach to resource management, involving the compiler and the run-time system. Program graphs based on the U-interpreter model is transformed into equivalent graphs which are more suitable for execution on the Tagged Token Dataflow Architecture, These graphs have predictable resource requirements and include special operations to engage the run-time system. The run-time system has two responsibilities: dynamic allocation/deallocation of resources, and dynamic control of program execution. The work presented here is motivated by the need to however, it serves a more general goal as well. Resource management is a fundamental aspect of any dataflow machine, and the issues raised in this thesis should have a prominent role in the design and evaluation of dataflow architectures in general.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1985
- Accession Number
- ADA154773
Entities
People
- D. E. Culler
Organizations
- Massachusetts Institute of Technology