In this part, the topic of performance evaluation shows those parameters that are devised to calculate the performances of various parallel systems. Achieving the highest possible performance has always been one of the major goals of parallel computing. Unfortunately, most often the real performance is less by a factor of 10 and even bad as compared to the designed peak performance. This creates parallel performance evaluation an area of priority in high-performance parallel computing. As we already know, sequential algorithms are mostly analyzed on the basis of computing time i.e., time complexity and this is directly related to the data input size of the trouble. For example, for the trouble of sorting n numbers using bubble sort, the time complexity is of O (n2). Though, the performance analysis of any parallel algorithm is dependent upon three main factors viz. time complexity, total number of processors required and total cost. The complexity is normally related with input data size (n).
Therefore, unlike performance of a sequential algorithm, the evaluation of a parallel algorithm can't be carried out without considering the other vital parameters like the total number of processors being employed in a definite parallel computational model. Thus, the evaluation of performance in parallel computing is depend on the parallel computer system and is also dependent upon machine configuration like PRAM, combinational circuit, interconnection network configuration etc. in addition to the parallel algorithms used for a variety of numerical as well non-numerical problems.
This unit gives a platform for understanding the performance evaluation methodology as well as giving an overview of some of the famous performance analysis techniques.