Large Grain Data Flow Graph Construction and Restructuring Utilizing the ECOS Workstation System.
Abstract
The U.S. Navy's new multiprocessor, the AN/UYS-2 Enhanced Modular Signal Processor (EMSP) utilizes a First-Come-First-Serve (FCFS) algorithm to transfer data. This algorithm is simple to implement but provides no mechanism to control execution of a specific application on the AN/UYS-2 which prevents performance predictions. A Large Grain Data Flow (LGDF) representation of a specific application is utilized to predict performance, with the introduction of trigger queues (dependency arcs) into the graphs to control execution. I utilized the EMSP Common Operational Software (ECOS) Workstation to execute graph representations of specific applications used by the U.S. Navy in the Anti-Submarine Warfare (ASW) arena. A complete description of the ECOS workstation, and the process of transforming specific applications into graph representations to be executed on the ECOS Workstation is demonstrated. Specifically, the Correlator Graph which represents a real-time ASW process is examined. To control and improve performance, the technique of implementing trigger queues using the ECOS Workstation is demonstrated. A basic graph is executed and referenced as a benchmark, with two reconstructed graphs executed demonstrating how trigger queues effect graph execution. The node execution times statistics indicate trigger queues control execution and will provide a mechanism to predict node performance.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1994
- Accession Number
- ADA289632
Entities
People
- Richard T. Keys
Organizations
- Naval Postgraduate School