On the Establishment of Standards for Comparing Algorithm Performance

One of the most frustrating tasks encountered by the OR practitioner is that of attempting to select an existing algorithm for application to an immediate real world problem. For purposes of illustration only, consider the OR analyst who is faced with having to solve a large zero-one integer programming problem. Being a member of ORSA, TIMS, SIAM, and other organizations, he realizes that a considerable number of publications have been addressed to this topic and thus he might be able to use one of these to solve the problem at hand. This is sometimes his first mistake.

Let us assume that he has surveyed the literature and found “n” publications presenting algorithms which deal directly with his particular problem. At this point in his search he will begin to discover that the mere existence of pertinent algorithms in the literature is not nearly sufficient.