Section: New Results

Quality of service and transport layer for future networks

On the Impact of the Flow-Size Distribution's Tail Index on Network Performance with TCP Connections

Participant : Paulo Gonçalves.

This is a joint work with Oana Goga (UPMC, Lip6) and Patrick Loiseau (Eurecom).

In this work, we studied the impact of the flow-size distribution on network performance in the case of a single bottleneck with finite buffer. To tackle the case where flows are transmitted with the TCP protocol, we use real experiments and ns-2 simulations. Our preliminary results show that the distribution's tail index impacts the performance in a more complex way than what is reported in existing literature. In particular, we exhibit situations where a heavier tail gives better performance for certain metrics. We argue that a main cause of our observed results is the transient behavior at the beginning of each flow.

Available Bandwidth Estimation for Multihop Wireless Networks

Participants : Isabelle Guérin Lassous, Van Nam Nguyen.

Estimating the available bandwidth in IEEE 802.11-based multi-hop wireless networks is a very difficult task due to the medium sharing among contending nodes and collisions between hidden stations. Several methods have been proposed so far for these networks to compute the available bandwidth on wireless links. If some recent solutions such as ABE and IAB now take into account collisions and their impact on the mean backoff, none of them considers the packet retransmissions due to collisions, although these retransmissions have an impact on the available bandwidth. In this work, we have proposed a new available bandwidth estimation for multi-hop wireless networks called RABE (Retransmission-based Available Bandwidth). This method integrates the average number of retransmission attempts in the available bandwidth estimation, in addition to other relevant parameters like the idle periods durations and the collision probability. RABE has been evaluated by simulation and the obtained results show that RABE can achieve a mean error ratio of 17% in comparison with the real measurement. Furthermore RABE is at least two times more accurate than ABE and ten times more accurate than IAB.

On The Recovery Performance of Single- and Multipath OLSR in Wireless Multi-Hop Networks

Participants : Inès Doghri, Isabelle Guérin Lassous.

In this work, we study and improve the recovery properties of single- and multipath routing strategies when facing network failure situations. In particular, we focus our study on two MANET routing protocols: OLSR and its multipath extension MP-OLSR. In various wireless multi-hop network environments, especially in multiple chain topologies, we define and evaluate the latency introduced by these protocols to find a new path after a link failure. Theoretical estimations and simulation results show that, under dual chain-topologies, this latency can be too long and incompatible with the needs of loss and delay constrained applications. As the source nodes cannot detect link failures immediately because of the delay incurred by the well-known nature of link state protocols in general, and of OLSR Topology Control (TC) messages in particular, these nodes keep sending packets along broken paths. We thus study the inconsistencies between the actual network topology and the nodes’ own representation. After analyzing the consequences of this long latency, we seek to alleviate these problems with the introduction of adapted mechanisms. We propose three new different schemes and accordingly extend the original OLSR and MP-OLSR protocols in order to decrease the expected latency and improve the protocol performance. Simulation results show a steep decrease of the latency when using these new schemes in dual chain-topologies. We also discuss these results in terms of packet loss, end-to-end delay and overhead.