A machine learning based approach for multi-domain service deployment in future networks

Abstract

The exponential increase of data traffic from new and improved services in healthcare, enhanced mobile broadband communication, transportation, augumented reality has led to cloudifing these services into an ordered set of virtual network functions known as Service Function Chains (SFC). These Service Function Chains(SFCs) ,deployed on remote cloud networks, are to be embedded onto the multiple domains belonging to different Infrastructure Providers(InPs) based on their resource requirements in relation to the available resources within the Infrastructure Providers (InPs) capacity while adhering to the stringent requirements of these services. The challenge lies when the Infrastructure Providers (InPs) are unwilling to disclose their internal topological information for security reasons which makes the effective allocation of resources complex. This calls for an intelligent algorithm that uses minimum information disclosed from the Infrastructure Providers (InPs) based on historical data to identify suitable domains for deployment. In this Project Work, a framework that deploys these Service Function Chains (SFCs) onto different domains that meet the delay constraints while at the same time respecting the privacy of Infrastructure Providers (InPs) is formulated. The partitioning of these Service Function Chain requests (SFC) is done by the proposed Reinforcement Learning Algorithm that identifies the most suitable candidate for deployment of the sub-SFCs. Simulation results, considering both online scenarios, reveal that the proposed algorithm results in up to 10% improvement in terms of acceptance ratio and provisioning cost compared to the benchmark algorithms, with up to more than 90% saving in execution time for large networks. In addition, an enhancement of the algorithm results in up to 5% improvement in terms of provisioning cost.