Solutions to Parallel and Distributed Computing Problems: Lessons from Biological Sciences

6.5 SEMISTATIC MATCHING AND SCHEDULING OF SUBTASKS

6.5.1 Overview

In this section, the application domain and HC platform assumed for the semistatic approach are described [7, 8]. This is followed by a discussion of the enhancements made to the GA described in Section 6.4 for determining off-line mappings in this situation. Finally, Section 6.5.7 summarizes the results of an extensive performance study of a simulated semistatic mapping system [35].

6.5.2 Application Domain: Automatic Target Recognition

One type of automatic target recognition (ATR) system takes a stream of images from a group of sensors and produces a scene description [54], tracking the movement of possible targets through the sequence of images. A simplified example of an ATR task can be found in [35]. The various types of image-processing elements required in an ATR system can be broadly classified into three groups: low-level processing (numeric computation), intermediate-level processing (quasi-symbolic computation, e.g., where numeric and symbolic types of operations are used to describe surfaces and shapes of objects in the scene), and high-level processing (symbolic computation, e.g., used to produce the scene description) [1, 4]. Each of these subtasks may allow the use of multiple processors of the same type for efficient execution. Heterogeneous parallel architectures consisting of multiple copies of different types of processors (e.g., SHARC DSP and PowerPC RISC processors [16]) are appropriate computing ...