not provide a full evaluation of multihoming and may lack a general applicability if based on a specific protocol.
Optimization techniques dealing with NP-hard problems cannot impose deploy- ment concerns or introduce more complexity on networks that are getting each day more complex to accommodate new services. Optimization can be formulated through distinct techniques, such as LP and MADM. Flexible techniques like MADM that do not require any adaptation between different scenarios are more interesting to future networks. Nonetheless, several issues are pointed to these techniques such as the ranking identification and ranking abnormality that compromise their accuracy and efficiency.
The main goal of this thesis is the proposal of an efficient and flexible optimization algorithm for multihoming that does not have the issues of MADM techniques. More- over, such algorithm relies on the different goals, above-mentioned, that multihoming solutions must pursue, namely resilience, ubiquity and load sharing. In addition, this thesis has also been motivated by a framework to evaluate multihoming support in different protocols, through the assessment of multihoming goals achievement.
1.2
Objectives and Contributions
The main goals of this thesis are to propose a mechanism to evaluate multihoming support and a mechanism to optimize the multihoming experience of a node with multiple interfaces or paths. The specific goals of the thesis include:
G.1 Enable objective evaluation of multihoming support, promoting the com- parison between protocols, regarding the efficiency of the multihoming mech- anisms.
G.2 Propose a mechanism that can optimize the multihoming experience of a multi-interface or multi-path node. The optimization mechanism must include the criteria that is employed in the multihoming evaluation mechanism and must be efficient and without introducing hard-to-meet requirements.
Several contributions are associated with this thesis, as summarized in the follow- ing subsections.
Multihoming Taxonomy
The multihoming concept has been objectively defined to avoid misunderstandings with related terms, such as multi-access. In addition, a state of the art has character- ized multihoming support in terms of goals and distinguished the diverse multihom- ing types, namely end-host, end-site and hybrid.
Multihoming evaluation framework
This contribution meets the G.1 goal, as a framework to assess multihoming support is proposed. This framework considers multihoming goals, more specifically resilience and ubiquity to determine how efficient a protocol is regarding its multihoming sup- port.
The Resilience Evaluation Framework (REF) and the Ubiquity Evaluation Frame- work (UEF) are frameworks that allow to assess resilience and ubiquity multihoming goals.
Performance assessment of multihoming evaluation framework
The performance assessment of Resilience Evaluation Framework (REF) and Ubiq- uity Evaluation Framework (UEF) was performed analytically. Diverse protocols have been employed as study cases. For instance, the Stream Control Transport Protocol (SCTP) [Eklund et al., 2009] has been studied regarding its resilience support with the primary-backup protection model included in the native specification of SCTP.
In a comparative approach, MIPv6 and HIP protocols have been assessed regard- ing their ubiquity support.
Multihoming aware optimization mechanism
MeTHODICAL is the optimization mechanism introduced in this thesis, to enhance the multihoming experience of nodes with multiple interfaces/paths. MeTHODICAL is a flexible optimization technique that enables optimal path selection by considering multiple criteria and with a low computational complexity. Moreover, MeTHODICAL follows a MADM approach, allowing users to specify weights for the diverse criteria in an objective way.
MeTHODICAL has been integrated in the architectures of Combating Fire with Multihoming and Mobility (CoFIMOM) and Trustworthy and Resilient Operations in
1.2 Objectives and Contributions
a Network Environment (TRONE) projects to enhance communications in fire-fighting scenarios and cloud environments, respectively.
Multiple Attribute Decision Mechanism accuracy evaluation framework
Outranking MADM techniques, such as DiA [Tran and Boukhatem, 2008] are often evaluated subjectively or using methodologies that cannot be applied generically. This thesis includes a MADM accuracy evaluation framework that is based on Design of Experiments (DoE) [Sandanayake et al., 2008] to allow the objective comparison be- tween different MADM techniques, relying on statistical properties, such as F-statistic or coefficient of determination, R2.
Performance assessment of multihoming aware optimization mechanism
MeTHODICAL has been evaluated analytically using data collected in real scenar- ios and has been assessed in a cloud testbed within the context of high-volume data transfers.
The performance assessment of MeTHODICAL was based on objective metrics that establish the correct ranking of paths and determine the required handover ra- tios. Moreover, MeTHODICAL has been compared objectively with similar tech- niques such as TOPSIS [Figueira et al., 2005] and DiA using the proposed MADM accuracy evaluation framework.
Implementation to enhance multihoming support
One final contribution that is associated to this thesis is a set of implementations to enhance multihoming support. In the TRONE project, an implementation of the op- timization algorithm has been incorporated in the TRONE architecture. Indeed, the optimization algorithm has been employed to reconfigure SCTP regarding the path to use, when multiple addresses are available in a cloud context. The mCoA++ is an implementation of the MCoA protocol, which extends the multihoming support of MIPv6. This contribution has not been left on a closed community, but instead it has been made public and available to the global research community.