| OCR Text |
Show 4 tectures in tightly coupled parallelism to network protocols in a networked environment such as a local area network of workstations (LAN). Normally the same communication medium is also used for synchronization in architectures where the computing nodes do not share a common clock. For instance, in Transputers [7] a communication link is used for data exchange and for synchronization by message passing, and in network parallel environments such as PVM [12] or Express [4] network messages are the primary means of synchronization. 1.3 Performance Measures Performance is an attribute that can be gauged both for the hardware platform and for the algorithm used. Performance observed while solving the same problem on different platforms can vary just as much as the performance observed for different algorithms being run on the same platform. Since hardware performance in general is a totally different subject, the following discussion will be limited to performance of algorithms. It should be noted, however, that these measures must always be qualified by the hardware platform on which the algorithm is implemented. Two obvious measures of performance for any algorithm are the storage requirements and execution time. Usually, the storage requirements have the least flexibility in terms of improving performance, and often the reduction of storage requirements will result in increased execution time and added complexity. The concurrency and scalability characteristics of parallel distributed algorithms add another twist as to how performance can be defined and measured. There is certainly a myriad of parameters that can be used as performance indicators depending on the architecture, configuration, the type of algorithm, etc., and possibly even the nature of the problem being solved. In this work two quantifiable measures of performance will be defined: parallel efficiency and speedup. Assume a problem of fixed size being solved in a parallel distributed environment using a certain algorithm. If the time it takes to sequentially solve the problem on one computing node is denoted by T8 and the time it takes to solve the same problem using a |
