tailieunhanh - Parallel Programming: for Multicore and Cluster Systems- P17

Parallel Programming: for Multicore and Cluster Systems- P17: Innovations in hardware architecture, like hyper-threading or multicore processors, mean that parallel computing resources are available for inexpensive desktop computers. In only a few years, many standard software products will be based on concepts of parallel programming implemented on such hardware, and the range of applications will be much broader than that of scientific computing, up to now the main application area for parallel computing | 152 4 Performance Analysis of Parallel Programs Performance Evaluation of Computer Systems The performance of a computer system is one of the most important aspects of its evaluation. Depending on the point of view different criteria are important to evaluate performance. The user of a computer system is interested in small response times where the response time of a program is defined as the time between the start and the termination of the program. On the other hand a large computing center is mainly interested in high throughputs where the throughput is the average number of work units that can be executed per time unit. Evaluation of CPU Performance In the following we first consider a sequential computer system and use the response times as performance criteria. The performance of a computer system becomes larger if the response times for a given set of application programs become smaller. The response time of a program A can be split into the user CPU time of A capturing the time that the CPU spends for executing A the system CPU time of A capturing the time that the CPU spends for the execution of routines of the operating system issued by A the waiting time of A caused by waiting for the completion of I O operations and by the execution of other programs because of time sharing. So the response time of a program includes the waiting times but these waiting times are not included in the CPU time. For Unix systems the time command can be used to get information on the fraction of the CPU and waiting times of the overall response time. In the following we ignore the waiting times since these strongly depend on the load of the computer system. We also neglect the system CPU time since this time mainly depends on the implementation of the operating system and concentrate on the execution times that are directly caused by instructions of the application program 137 . The user CPU time depends both on the translation of the statements of the program into