In this article, the author discusses the importance of benchmarking when measuring the performance and capacity of a program. They explain that traditional methods of measuring CPU usage are not enough and introduce the concept of the response-time curve, also known as the “hockey-stick curve.” This curve shows the point at which the program’s performance starts to decline and can help determine the program’s bottleneck. The author also provides examples and tools that can be used to estimate the performance and capacity of a program. They emphasize the need for capacity planning and highlight the advantage of using queuing network models for this purpose.
https://queue.acm.org/detail.cfm?id=3595862