SLAs Decomposition for Network Slicing: A Deep Neural Network Approach

5G is a multi-service network supporting a range of verticals with a diverse set of requirements. Network slicing enables the creation and operation of multiple logical networks over the shared network infrastructure, tailored to the requirements of a particular service type with agreed upon Service Level Agreement (SLA). A network slice may span different parts of the network (i.e., access, core, and transport network) and could be deployed across multiple operators and infrastructure providers. Therefore, the end-to-end SLA associated with a network slice should be decomposed in portions attributed to each of these domains. We assume a two-level management architecture consisting of an end-to-end service orchestrator responsible for the lifecycle management of the network service and domain controllers that are in charge of instantiating parts of the slice in their respective domains. In this context the orchestrator is responsible for decomposing the SLA, without detailed knowledge of the state of the resources in each of the domains. The orchestrator is only aware of the responses of the domain controllers to previous requests and captures this knowledge in a risk model associated with each domain. In this study, we propose an approach for decomposing the end-to-end SLA adopting neural network-based risk models. We discuss several approaches that utilize the monotonicity prior, such that the SLA can be adequately decomposed even when the number of historical data is low. An empirical study on a synthetic multidomain dataset demonstrates the efficiency of our approach.