| Abstrakt: |
The problem of designing communicating systems with internal state and facilities for reconfiguration is considered. A technique for designing such systems, called dynamically reconfigurable systems, based on a data driven model is proposed. The data driven model uses undirected or partially directed graphs. The nodes of the graphs can represent the components of a concurrent system, modules of an operating system including local and global control elements, and modules of a knowledge-based expert systems. The arcs can represent interconnections between components or data paths for intermodule communication. Each node has a number of attributes including a distinct name, an operation, and a set of specifications describing its input and output arcs. Each arc also has a number of attributes including an arc name, arc type, and the maximum number of tokens the arc can carry. Each node has local memory to store the values of the node and arc attributes and executes in data driven manner. A configuration is an assignment of values to the attributes of the nodes and arcs. Special nodes, called configuration specifying nodes, can assign values to the attributes of nodes and arcs. This process is termed reconfiguration. Reconfiguration is based on the information gathered from the nodes using “snoop arcs.” A formal description of the technique and the execution of a graph are described. The issues related to what, when, and how to reconfigure are identified and resolved. Several reconfiguration strategies are presented using local information, global information, and a combination of the two. Problems arise when local and global requirements are in conflict. A conflict free approach to reconfiguration is presented. Dynamic systems that cannot be accurately modeled using Petri Nets can be modeled using the proposed technique. For example, the Cray-1S computer system has been modeled using the proposed methodology and its architecture analyzed to improve performance. The proposed improvements have been incorporated in the Cray-XMP computer. |
Full text from SpringerLink