Contributions to the algorithms and methodologies for virtual infrastructure provisioning
- Autores: Jordi Ferrer Riera
- Directores de la Tesis: Xavier Hesselbach (dir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2016
- Idioma: inglés
- Tribunal Calificador de la Tesis: Carlos Bock Montero (presid.), Sebastià Sallent Ribes (secret.), Emmanouil Kafetzakis (voc.)
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
Virtualization is defined as the process consisting of building a virtual resource on top of one or several physical resources, depending on the selected paradigm: aggregation or partitioning. From the net- working perspective, over the recent years virtualization became a hot topic in networking research, although the term is not new, neither for networking nor for Information Technology (IT). In virtualization environments, the most significant challenge to be addressed is the virtual infrastructure allocation or embedding one. The problem appears over any networking technology, from wired to wireless domains, including converged physical infrastructures as well as coupled IT and networking infrastructures. This problem has re-gained importance in the last years due to the emergence of novel software- based abstractions, which provide fully-operational interfaces on top of those virtual resources, enabling the emergence of novel business models. The election of the physical resources upon which the virtual infrastructure will be instantiated becomes fundamental for the virtual infrastructure operation in those virtualized environments, i.e. the election of the physical resources over which the virtual ones will be allocated. In fact, the virtual infrastructure allocation system might be seen as a policy-based system, where, based on the physical infrastructure available, and the set of requests, it will create different virtual infrastructures as a function of the policy, which will indicate the metric to optimize when creating those Virtual Infrastructures (VIs). For example, the infrastructure provider, who is responsible for leasing those virtual infrastructures, may want to optimize the economical cost of the different instantiated virtual slices, or even the overall energy consumption of the slices. The problem acquires different dimensions when considering heterogeneous resources (e.g. optical network resources and wireless resources) in order to create cross-domain virtual infrastructures; or when it includes IT resources connected at the edges of the different network domains. The first part of the thesis provides an insight on the problem, and different contributions on the algorithms utilized in order to solve the problem. It considers different optical substrate technologies, wireless and wired resources, and it tries to optimize either the total number of virtual infrastructures to be created or either the energy consumed by the resources. Furthermore, the work proposed in the second part of the dissertation includes a dynamic re-planning approach, which is responsible of dynamically updating already provisioned virtual infrastructures in order to adapt them to the Cloud-based services. Those services are unpredictable; basically, information regarding the volume and type of service requests is not precisely available in advance of the requests to the VI providers. Cloud services can scale up and down on demand. Therefore, that dynamic adaptation of the infrastructure to the elasticity of the Cloud services requires constant changes. Dynamic re-planning becomes fundamental for virtual infrastructures supporting cloud-based services. Virtualization can be indistinctly applied to IT servers, network re- sources, and even network functions. The last part of the work de- fines the additional scheduling problem, specific for Network Function Virtualization (NFV), which is complementary to the allocation or embedding. NFV aims at decoupling the network functions from the actual hardware where they are executed. Thus, most of the network functions will run on standard virtualized servers by means of soft- ware, which is expected to significantly reduce capital expenditures, while at the same time the software automation reduces operational expenditures. The solution for the scheduling aims at minimizing the total execution time for chains of network services running over IT infrastructures.
