||As an illustrative example, this demo considers an SDC with two (UK and U.S.) domains, which allocate analytic tasks to data servers for processing. Each domain has a local reinforcement-learning (RL) agent to distribute processing requests (tasks) to servers within the domain. When the SDC is formed, a global RL agent is established to interact with the two local agents so that processing requests can now be allocated to servers anywhere on the SDC (i.e., possibly across the domain boundary) for efficiency. The purpose of this demo is two-folded. First, training the local RL agent is challenging due to the space-action explosion. We implement and show the recently developed technique that separates state space from action space  can enhance training. Second, this demo shows that a transfer-learning (TL) approach  to training the global RL agent can reduce the time required for achieving close to the optimal performance after the SDC is formed or the domains are reconnected after fragmentation. By combining the state-action separation technique for the local agents as well as the TL approach for the global agent, this demo shows that SDC fragmentation and reconnection can be handled intelligently and efficiently.