CBR and Workflows for e-Science

I was talking with my friend Jaliya about an interesting application of CBR (Case-Based Reasoning) to workflows in e-Science.
Most of the eScience systems that use workflows, run these workflows on super computers. These operations might take a considerable amount of time even within super computers and the overhead of the system itself might degrade the performance of a system. Sometimes we might have to run multiple workflows to get a good decision or output.
If we take weather applications for example, we will run couple of workflows to get a final decision. What if we can look at the data, compare them with previous data and results, and then predict the workflows to run? It is true that no two weather events are similar (a quote from my friend Suresh Marru). But this method might give really fast clues to those meteorologists.

We were trying to do this as a class project, but failed as we couldn't find enough data within the system. But for me this seems a very promising approach.

During one of our extreme lab brainstorming coffee session, we thought about the application of CBR for scheduling jobs within grid environment. If we have data like, process requirements (# of CPUs, time constraints, etc.,) and previous information like wait time, execution time, affinity for each super computer, then we can build a AI system, with concepts from CBR. An interesting project to do, if I get some time.