Conception et implementation d'un gestionnaire de flux SIP

SIP is widely used by the telecom industry. This implies that the number of users using related services is very high. Thus, QoS requires that SIP network components remain functional at all times. However, traffic burst usually harms the network performance because main components such as SIP server could be overloaded. UDP retransmissions usually exacerbate the situation. As a result, more calls are rejected and signaling delays increase drastically.;SIP sessions are established based on message exchange. Each message has a different tolerance to delay and rejection. Message processing should depend on the message type through scheduling and message priority. SIP network should possess mechanisms capable of managing SIP message conveniently.;A new approach of SIP message management has been introduced in order to improve SIP network performance in case of traffic burst. A differentiated handling of SIP messages has been implemented in a GFSIP in order to process SIP requests and responses differently. Different tolerances have been established for the same SIP session. Three queuing disciplines have been implemented to study their effect on delay and call rejection. The study of queuing discipline in this context was challenging because SIP signaling is organized into sessions.;The diagnosis of SIP performance was carried out through a SIP test-bed. Message exchange between SIP entities has been illustrated via a proposed scenario. Markov Models have been used to illustrate signaling steps and their relationship with the tolerance. During tests, bursts targeted the GFSIP and the SIP server was protected. Analyses were conducted based on delays and rejection rates variation. The result demonstrated the effect of different queuing disciplines on QoS.