Development of a Protocol Usage Guideline for Conservative Parallel Simulations

Abstract

The objective of distributed simulation is to speedup simulation execution by partitioning the simulation processing load over multiple processors, This thesis reviews current synchronization protocol methods for distributed simulations, and proposes guidelines for obtaining optimal conservative simulation partitionings using empirical evidence. An analysis is performed using three protocol variations of the Chandy-Misra HULL message algorithm, two using a pending message blocking strategy, and the other using a safetime blocking strategy. A fourth protocol evaluated is based on the SRADS algorithm proposed by Reynolds. The analysis involves a study of all possible 2 and 4 node configurations, for three queuing simulations, using all possible protocol and model pairings. A fourth queuing model is then used to independently validate results. In the end, the safetime version of the Chandy- Misra protocol is demonstrated to provide better overall performance than the other protocols evaluated. Partitioning guidelines developed established a relationship between process configurations and load balancing. It is seen that separating highly communicative processes onto different nodes, or locating highly communicative processes on the same node with fewer processes, provided the optimal 4 node configurations, while reducing the number of intra-node process connections provided for the optimal 2 node configurations.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 01, 1992
Accession Number
ADA258851

Entities

People

  • Prescott J. Van Horn

Organizations

  • Air Force Institute of Technology

Tags

Communities of Interest

  • Engineered Resilient Systems
  • Human Systems
  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Abstracts
  • Air Force
  • Algorithms
  • Application Software
  • Central Processing Units
  • Computations
  • Computer Simulations
  • Computers
  • Experimental Data
  • Hybrid Simulations
  • Instrumentation
  • Measurement
  • Operating Systems
  • Schematic Diagrams
  • Simulations
  • Standards
  • Time Intervals

Fields of Study

  • Computer science

Readers

  • Computer Networking
  • Parallel and Distributed Computing.