As you might expect, there are special benchmark tests that strain OLTP systems to see how well they work in particular situations and how they compare to one another. OLTP benchmarks differ from the benchmarks used elsewhere in the computing industry because, for example, OLTP systems often value data-access speed more highly than the raw computational power that's valued in, say, scientific modeling applications. The trouble with OLTP benchmarks is that every OLTP application is different. The system that backs up a network of automated teller machines has a different set of demands placed upon it than the system behind the NASDAQ stock market.
When benchmarking an OLTP system, choosing the correct tool is imperative. The book includes a collection of technical papers on several different benchmarks, including the popular TPC-A, TPC-B, and TPC-C benchmarks, and shows the strengths and weaknesses of each. The Benchmark Handbook for Database and Transaction Processing Systems concludes with a discussion of designing your own benchmarks. You'll want to get this book if you're interested in figuring out how to test a system's performance.