Document Type

Technical Report

Publication Date

Fall 1-1-1975


In this paper, some simulation design techniques used in the study of a new architecture called the multi associative processor (MAP) architecture are discussed. The architecture executes multiple single-instruction-multiple-data stream (SIMD) programs simultaneously (i.e., it is a specialized multiprocessor). The design analysis of the system includes topics such as the generation of a suitable job mix for the machine; simulating the execution of a SIMD program; and simulating the competition between executing programs for various shared resources. The approach to analysis of the MAP system design is based on a highly detailed (low level) simulation model that interpretively executes a single SIMD program at a time. This interpreter provides a medium for testing various SIMD algorithms that could be executed on a real system corresponding to the hypothetical MAP system. Although the interpreter does not allow for the simulation of the simultaneous execution of two or more SIMD programs, the single executions are measured with a (simulated) hardware or software monitor to obtain data indicating the performance of that program in an isolated (non-competitive) environment. These data are then used to drive higher level (lower detail) simulation models that focus on particular aspects of the design, e.g., the competition for main memory cycles.