Proposta de Tese de Ricardo José Pfitscher

Aluno: Ricardo José Pfitscher
Orientador: Prof. Dr. Lisandro Zambenedetti Granville

Título: Monitoring and Identifying Bottlenecks in Virtual Network Functions Service Chains
Linha de Pesquisa: Redes de Computadores

Data: 20/12/2017
Horário: 13h30min
Local: Prédio 43412 – Sala 215 (sala de videoconferência), Instituto de Informática

Banca Examinadora:
– Prof. Dr. Guilherme Piêgas Koslovski (UDESC via videoconferência)
– Prof. Dr. Jéferson Campos Nobre (UNISINOS)
– Prof. Dr. Luciano Paschoal Gaspary (UFRGS)

Presidente da Banca: Prof. Dr. Lisandro Zambenedetti Granville

Abstract: Network functions perform an important role to the communication between clients and end-services in computer networks. These functions, traditionally provided by hardware-based appliances, are employed to achieve various objectives, such as security, addressing, routing, caching, and load balancing. The virtualization trend that drove cloud computing has also been applied to network functions, resulting in the Network Functions Virtualization (NFV) concept. NFV implements hardware-based network functions (a.k.a middleboxes) as Virtual Network Functions (VNFs); with this paradigm shift, network providers benefit from: capital and operational expenditure reduction, acceleration of the time-to-market of novel solutions, and on-demand resource scaling. One of the main features of NFV is that VNFs can be chained and provisioned on demand, bringing elasticity and dynamicity to the network. In particular, Service Providers take advantage of forwarding graphs to specify service chains and provide customized network services to meet individual client demands. An implication of the  interdependency between VNFs is that resulting service chains may not work as expected, and because of that, it is crucial to determine which VNFs are having a negative impact on the service quality. Usually, network operators detect bottlenecks and quantify performance degradations by monitoring both virtual and physical resources, and through analytical modeling of the various components involved in queuing networks communication. However, NFV imposes additional characteristics that hinder the adoption of these approaches: the heterogeneity of environments with all sorts of VNF purposes (e.g. provide security, address translation, and performance optimization), and the unusual functioning of specific VNFs that rely on non-blocking I/O implementations. In this thesis, a model to quantify the guiltiness of a VNF on being a bottleneck in a service chain is proposed. Such model consists of a weighted sum of relevant metrics found in the literature, refined through numeric assessments on representative scenarios. In addition, an adaptive algorithm based on linear regression and neural networks is introduced to adjust the model parameters according to the environment particularities, such as the type and number of VNFs. Experimental evaluation were conducted in representative scenarios to assess the ability of the model in i) detecting bottlenecks, and ii) quantifying performance degradations. Results show that the guiltiness metric faithfully characterizes end-service performance, identifying up to 94\% of the bottleneck VNFs in the analyzed scenarios, and estimating end-service response time with a 0.98 R-squared precision. Furthermore, studies are carried out to establish the appropriate method for monitoring NFV environments. To this end, a Distributed Result-Aware Monitoring (DReAM) approach is proposed: a scalable monitor that can achieve near real-time bottleneck detection.

Keywords: Network Functions Virtualization, Bottlenecks Identification, Performance Analysis.