Autonomic resource partitions for adaptive networks

A method for autonomous resource partitioning in a network, the method comprising:

• provisioning a plurality of resources including any of virtual resources and physical resources in one or more layers in the network and monitoring availability of the plurality of resources in the network;
• determining which of the plurality of resources are required for Quality of Service (QoS) in the network and satisfying the QoS required recursively using resources of the plurality of resources provided by a layer below and resources of the plurality of resources provided at a layer thereof requested by a partition manager;
• advertising and assigning resource requests to corresponding resources recursively to resource partitions in a layer above;
• and
• tracking the utilization of the resources provided by the one or more layers and adjusting resource usage of the resources based on a negotiation to minimize a cost of implementation wherein one or more of the partition manager and the resource broker utilize a deep-Q network (DQN) in the negotiation that includes reinforcement learning and deep neural netwroks, wherein an operator specifies rewards and network costs to converge the plurality of resources in the network to a desired state.

A reinforcement learning system for autonomous resource partitioning in a network, the reinforcement learning system comprising:

• one or more processors;
• and
• memory storing instructions that, when executed cause the processor to implement building blocks of layered virtual functionality including:
• a resource controller configured to provision a plurality of resources including any of virtual resources and physical resources in one or more layers in the network and monitor availability of the plurality of resources in the network;
• a resource broker configured to advertise and assign resource requests to corresponding resources of the plurality of resources recursively to resource partitions in a layer above the resource broker;
• and
• a partition manager configured to track the utilization of the resources provided by the one or more layers and to adjust resource usage of the resources in negotiation with the resource broker to minimize a cost of implementation,
• wherein one or more of the partition manager and the resource broker utilize a deep-Q network (DQN) in the negotiation that includes reinforcement learning in the negotiation that includes reinforcement learning and deep neural networks, wherein an operator specifies rewards and network costs to converge the plurality of resources in the network to a desired state, wherein the partition manager receives resource partition state information and provides resource adjustment information to the resource broker, and wherein the resource broker provides resource costs to the partition manager and commands for resource adjustments to the resource controller.