On the Orchestration of Dynamic Services over Distributed IT Infrastructures
The arise of Network Function Virtualization and Software Defined Networking enabled the programmability of the network infrastructure, hence allowing the dynamic deployment of complex services on top of the network. Additionally, Edge Computing extends the Cloud paradigm towards the edge of the network, leading to a highly distributed infrastructure and introducing new players in the service provisioning ecosystem. This transformation leads to enhanced possibilities for service providers, which, thanks to the flexibility and new capabilities offered by the infrastructure below, may realize and deliver a plethora of new applications, such as virtual reality, remote critical tasks, and more. However, managing such a distributed infrastructure and enabling interoperability between the multiple actors involved introduces a series of challenges.
A particularly challenging problem is resource management. Since resources (such as computing, networking, and storage) should be partitioned in slices that are allocated for each service, a key component called Orchestrator is often employed to decide on the deployment and the management of each service. However, the optimality of the taken decisions may not match the actual necessity of the services, as each of them may benefit from different allocation strategies and may want to optimize on different parameters and service-specific metrics. Such metrics are often unknown to the orchestrator, which operates at the infrastructure level and based on a one-size-fits-all paradigm. Moreover, mandating the existence of centralized coordination components may not be suitable in a scenario where services are executed on scattered compute nodes, e.g., at the edge of the network, which features arbitrary and dynamic topologies. Finally, since resources are scarce and geographically distributed in different areas, service provisioning may involve multiple providers that should inter-operate coordinating the deployment of applications on top of their clusters.
Given these considerations, this thesis investigates new service-centric orchestration paradigms, which cover different aspects of the above problem. A novel Service-Defined Orchestration approach is proposed, which distributes the orchestration task and delegates it to the service providers competent for each application. The problem of service management, utilization, and dissemination in harsh environments is also investigated, by designing highly distributed architectures and algorithms with the aim of transparently enable a suitable service layer on heavily scattered IT infrastructures.
Department: PhD in Computer and Control Engineering
Funded by: Tierra Telematics
Supervisor: Prof. Fulvio Risso