ABSTRACT
If one is trying to buy a new car or a new super-computer, one can run a series of standard tests to distinguish which is the best model to buy, which gives best value for money, and which should not be touched with the proverbial bargepole. This sort of inter-comparison goes under the general heading of “benchmarking,” a term quite literally derived from carpentry, whereby carpenters measure one piece of wood against another on their workbench by means of a series of marks—bench-marks. One test for a car might be the time it takes to accelerate from a standing start to 100 kilometres per hour. For a computer one might have a series of programmes designed to get behind the manufacturers’ claims of gigaflops per processor and find out the actual time it would take to run the job(s) that needed running—conceptually straightforward, if not necessarily straightforward in practice.
