Load test starts at a minimum number of threads or requests per second and increases by a given increment to a maximum number of threads or requests for a given duration then decreases until it reaches the defined minimum.
(Graphically this could be represented as a bell-curve.)
So for instance if the test was to run for one hour in might start with a minimum of ten requests per second and increase the number of requests every five minutes to reach the maximum number 100 of requests after twenty minutes , run at maximum for another twenty and then decrease the number of requests until it falls to the minimum. Ideally it would be best to be able to specify a lead-in and lead-out period so that the test could run at the minimun level for a given duration before the rate of request / no of threads begins to increase and after it falls back to the minimum level.
Build up gradually, run a t maximum for a given duration and fall off to the minimum level suddenly.
Rapid climb and rapid descent.
Run at maximum level then taper - off to the minimum.