No abstract

No abstract

No abstract

Ad Hoc networks are characterized by fast dynamic changes in the topology of the network. A known technique to improve QoS is to use Multipath routing where packets (voice/video/…) from a source to a destination travel in two or more maximal disjoint paths. We observe that the need to find a set of maximal disjoint paths can be relaxed by finding a set of paths S wherein only bottlenecked links are bypassed. In the proposed model we assume that there is only one edge along a path in S is a bottleneck and show that by selecting random paths in S the probability that bottlenecked edges get bypassed is high. We implemented this idea in the MRA system which is a highly accurate visual ad hoc simulator currently supporting two routing protocols AODV and MRA. We have extended the MRA protocol to use multipath routing by maintaining a set of random routing trees from which random paths can be easily selected. Random paths are allocated/released by threshold rules monitoring the session quality. The experiments show that: 1) session QoS is significantly improve, 2) the fact that many sessions use multiple paths in parallel does not depredate overall performances, 3) the overhead in maintaining multipath in the MRA algorithm is negligible.

When studying scalability in ad-hoc networks, most works present experimental results for a limited number of nodes (100-200). Various "explicit" clustering techniques have been proposed to improve scalability, obtaining successful sessions for 400-800 nodes. However, explicit clustering may damage the performance, e.g., cause session breaks due to fast movements of cluster heads. An alternative to explicit clustering is the use of algorithms that are "naturally clustered", i.e., arrange the nodes in dynamic hierarchical structures. In this work, we study the effect of explicit clustering by comparing an advanced version of the Ad Hoc Distance Vector Algorithm (AODV) with the Metrical Routing Algorithm (MRA) that possesses the natural clustering property. We cover fundamental aspects of scalability and experimentally prove the superiority of implicit clustering over explicit clustering. In particular, we consider heterogeneous theaters with several types of transmitters including personal, car-mounted, helicopters and a Geostationary (GEO) satellite. Natural clustering is more effective in heterogeneous theaters as the more powerful transmitters can serve as cluster heads. A formal bound based on general probabilistic assumptions shows that all existing ad-hoc algorithms cannot scale infinitely, thus rendering scalability as an experimental issue.