Resumen de Contribution to the improvement of the performance of wireless mesh networks providing real time services
Nowadays, people expectations for ubiquitous connectivity is continuously growing. Cities are now moving towards the smart city paradigm. Electricity companies aims to become part of smart grids. Internet is no longer exclusive for humans, we now assume the Internet of everything. We consider that Wireless Mesh Networks (WMNs) have a set of valuable features that will make it an important part of such environments. WMNs can also be use in less favored areas thanks to their low-cost deployment. This is socially relevant since it facilitates the digital divide reduction and could help to improve the population quality of life. Research and industry have been working these years in open or proprietary mesh solutions. Standardization efforts and real deployments establish a solid starting point.We expect that WMNs will be a supporting part for an unlimited number of new applications from a variety of fields: community networking, intelligent transportation systems, health systems, public safety, disaster management, advanced metering, etc. For all these cases, the growing needs of users for real-time and multimedia information is currently evident. On this basis, this thesis proposes a set of contributions to improve the performance of an application service of such type and to promote the better use of two critical resources (memory and energy) of WMNs.For the offered service, this work focuses on a Video on Demand (VoD) system. One of the requirements of this system is the high capacity support. This is mainly achieved by distributing the video contents among various distribution points which in turn consist of several video servers. Each client request that arrives to such video server cluster must be handled by a specific server in a way that the load is balanced. For such task, this thesis proposes a mechanism to appropriately select a specific video server such that the transfer time at the cluster could be minimized.On the other hand, mesh routers that creates the mesh backbone are equipped with multiple interfaces from different technologies and channel types. An important resource is the amount of memory intended for buffers. The quality of service perceived by the users are largely affected by the size of such buffers. This is because important network performance parameters such as packet loss probability, delay, and channel utilization are highly affected by the buffer sizes. An efficient use of memory for buffering, in addition to facilitate the mesh devices scalability, also prevents the problems associated with excessively large buffers. Most of the current works associate the buffer sizing problem with the dynamics of TCP congestion control mechanism. Since this work focuses on real time services, in which the use of TCP is unfeasible, this thesis proposes a dynamic buffer sizing mechanism mainly dedicated for such real time flows. The approach is based on the maximum entropy principle and allows that each device be able to dynamically self-configure its buffers to achieve more efficient memory utilization. The proper performance of the proposal has been extensively evaluated in wired and wireless interfaces. Classical infrastructure-based wireless and multi-hop mesh interfaces have been considered. Finally, when the WMN is built by the interconnection of user hand-helds, energy is a limited and scarce resource, and therefore any approach to optimize its use is valuable. For this case, this thesis proposes a topology control mechanism based on centrality metrics. The main idea is that, instead of having all the devices executing routing functionalities, just a subset of nodes are selected for this task. We evaluate different centralities, form both centralized and distributed perspectives. In addition to the common random mobility models we include the analysis of the proposal with a socially-aware mobility model that generates networks with a community structure.
