Complete course notes

Pagina 1

1 Lesson 2 28.02 Performance There are two ways to improve performances: • Improve the technology, by enhancing the clock speed or reducing the size of transistors, it is slow and not so important for this course; • Improve the computer architecture, which is the cause of all the major changes done in computers in the last 20 years. To have better performances it is important to follow a specific method. - Evaluate the existing system and identify the bottleneck, - Benchmarks, - Simulate new design and organization, - Implement the new system and commercialize it. These actions build a circle, so after the last step it is important to go to the first one to continually improve the system. Performance can be intended in two ways 1. Bandwidth or throughput, which represents the total work done in a certain amount of time 2. Latency or response time or execution time, which is the time between the start and the completion of an event. The definition which has to be considered depends on the focus that is needed: a data-warehouse manager will use the first one, while those who programme for single user will use the second one. These two points of view lead to two different concepts of “faster”, as saying that a computer is faster than another depends on what is important. So, two metrics can be used: 1. When the aim is to minimize the elapsed time for program execution, we have to consider the execution time, which is calculated as the difference between the end and the start of the program; 2. When considering the total amount of work, it is important to consider the maximum completion rate, which is obtained as number of jobs/seconds. The throughput is the reciprocate of the average response time only if there is no overlap, which means no concurrency; otherwise the…