See rand.

Also, it's good practice to delay every next process by at least as much as the previous request took; see time for getting the current time to determine the start and end time of a request.