- Probably can swap all malloc/memset pairs with calloc or some zmalloc sort of call.
- Code is tightly coupled with jupiter topology (especially in the planning "plan.h" and "config.h", and other places) Relying on the fact that a alocated switch is a switch from the jupiter topology. Have to fix it by relaying the calls to the network_t structure.
-
Adding Dynamic programming to the long-term planner
- Dynamic programming (Reward is going to be n-step later)
- Anytime that you have DP you have reinforcement learning
-
If you can provide some structure to your plans, then you don't need to search them both
- Testing just static traffic to flesh out the searching strategy