I'm writing a task-set generator tool for generating schedulable task-sets. The motivation behind the tool: After implementing a new spinlock protocol for a specific RTOS, the problem was testing the implementation. There wasn't a real (customer) application that the "company" could provide so that we could perform some evaluations against a baseline. So we developed this tool which is able to generate schedulable task sets and we used this tool to compare the performance of the state-of-the-art and our implementation. The tool generates a task-set with a given number of tasks such that each task has a well-defined execution time and period.
To generate a load for each task in the generated task set, we simply used the "sum of first n-integers". On my hardware platform, we first calculated the time it takes to sum, let's say, first 500 integers and then using this as my "unit runnable", we scaled up or down the "n" according to the generated execution time.
In the current version of our task-set generator, we use only one type of load generation function, i.e., waste CPU cycles. The good thing about this "class of load generation function" is that we can specify the number of iterations based on a "unit runnable". But this class of functions is not suitable to cover various types of behavioral characteristics.
So my question is, what other classes of load functions exists that can be used? I'm looking for functions for which I can define a "unit runnable" and then scale it up or down. This is so because the generator generates a specific execution time of the task which makes the task schedulable. The load function must be controllable so that we can respect the generated execution time.