A microbenchmark attempts to measure the performance of a "small" bit of code. These tests are typically in the sub-millisecond range. The code being tested usually performs no I/O, or else is a test of some single, specific I/O task.
Microbenchmarking is very different from profiling! When profiling, you work with an entire application, either in production or in an environment very painstakingly contrived to resemble production. Because of this, you get performance data that is, for lack of a better term, real. When you microbenchmark, you get a result that is essentially fictional, and you must be very careful about what conclusions you draw from it.
Still, for either type always apply the old adage:
Premature optimization is the root of all evil.
