Since OurGrid v4.0, when the WQR heuristic calculates the number of needed workers for a request, it uses the number of tasks of the job.
Another approach would be the number of tasks times the number of max replicas, which is more aggressive, but can reduce the make span.
The problem with this approach is because its advantages does not scales. If the job has 1000 tasks and the max replicas is 3. The makespan will not change too much if the broker receives 1000 or 3000 workers.
We need to simulate some scenarios and choose what is the best approach. Perhaps a function of the number of tasks.
Moving this issue to the Features tracker.