Contribución al estudio y diseño de mecanismos avanzados de servicio de flujos semi-elásticos en internet con garantías de calidad de servicio extremo a extremo

Résumé

Internet services and applications are increasingly demanding appropriate levels of service for their adequate operation. That supposes the development of new mechanisms to allow end-to-end differentiated quality of service. In these networks, the cost of using this differentiated service is consequently higher than the employment of the well-known best-effort delivery mode, which requires users to make a responsible use of the reserved resources. Therefore, it is essential to reserve only those resources strictly necessary to achieve the minimum cost. This Thesis work deals with the problem of minimizing reserved resources in a semi-elastic flow transmission. This type of flow is characterized for requiring the reservation of more or less resources depending on network conditions, and for this reason, it is considered the design of a client-server system to reserve automatically the required demanded resources. Hence, it is proposed a method to control the client memory occupancy that determines when it is necessary to use a best-effort delivery mode (those periods when occupancy guarantees data availability in the client side), and when to use a resource reservation delivery mode to guarantee enough data entering into the client memory. From the analysis of this method, it is deduced the transmission cost, which depends on several parameters directly involved with client memory occupancy. Among these parameters, the delivery data rate and the maximum occupancy threshold, that indicates when it is possible to send data using best-effort mode, are the most significant ones. Thanks to the correct dimensioning of this maximum threshold, the client is able to minimize reserved network resources. To calculate this threshold, it is necessary to acquire an initial estimate for the best-effort delivery rate. To achieve this, it is proposed the use of an initial threshold to force a change to best-effort mode to obtain this estimate. In addition, it is proposed the use of a mechanism to improve the minimization cost for high delivery rate variability, without excessively increasing the signalling associated to the switching of transmission modes. Finally, it is proposed a method to serve simultaneously various clients. A homogeneous case is studied by simulation, in which all clients have the same requirements. This method allows a particular number of clients to be served by reserving resources, while the rest use best-effort mode. From this study, it is determined the existence of an optimum value for the number of simultaneous clients to be served by using resource reservation, that minimizes extra signalling to manage this service. Furthermore, as an aid to complement this research work, it has been developed an implementation of a client-server system for semi-elastic flows using ns-2, and additionally, a real implementation using C programming language in a Linux system.