nsralloc-ml context & background

NSRAlloc-ML project summary and context

The Internet of things (IoT), 5G networks, edge/fog computing, vehicular networks, and cloud computing are systems with challenging requirements for modeling, orchestrating, and allocating resources in networks.

In this context, there is a need for dynamic network infrastructures that adapt to the demands and requirements of users. This adaptive capacity of the network commonly uses machine learning algorithms to deal with the inherent complexity of the problem and support the operational dynamism of the network. These new network structures are called intelligent networks, as a reference to their operational intelligence, or self-driving networks, as a reference to their ability to adapt to the operation.

The virtualization of network resources is another aspect of research with significant impact and relevance. In this case, the network resources are sliced and grouped to create virtual networks using shared physical resources with high cost, efficiency, and scale gains for companies and institutions.

The NSRAlloc-ML (Network Slicing Resource Allocation ML-Enhanced) project proposes the modeling and deployment of network slicing strategies to allocate communication resources capable of supporting the dynamism and elasticity of end-to-end communication users. NSRAlloc-ML uses new resource allocation models with bandwidth allocation models (BAM) aided by Q-Learning and SARSA algorithms for resource slicing and bandwidth allocation in networks.

The DyRA framework (Framework for Dynamic Resource Allocation) integrates the components of NSRAlloc-ML to provide a sliced, dynamic, and intelligent resource allocation over a physical network infrastructure. Intelligent bandwidth allocation in NSRAlloc-ML uses, in addition to machine learning algorithms, two other innovative elements: the SDN/OpenFlow paradigm for programming the network infrastructure and the Publish/Subscribe (Pub/Sub) strategy for a distributed data access scenario.

A dynamic, intelligent, and self-regulating resource allocation for network slicing is the target of the NSRAlloc-ML. It is an innovative solution to the NP-hard class of network slicing problems with communication resource allocation for 5G, telecommunication networks, and Edge/Fog/Cloud computing in various application scenarios.