Efficiently using
the network resources of Mobile Ad-hoc NETworks
(MANETs) is challenging. The absence of a centralized administration leads to a
congestion problem (Transport layer). The ows are
usually routed through shortest routes, typically through the same central part
of the network (Network layer). Communicating via shared wireless links raises
a contention problem (MAC layer). Multi-hop transmissions cause flows not only
to interfere with each other, but also with themselves.
We focus on
jointly solving the contention and congestion distributed control problem in a
bounded queue MANETs. The resulting ow rates satisfy fairness criteria
according to a given Network Utility Maximization (NUM) function. In recent
years a number of papers have presented solutions to the same problem based on
NUM algorithms. However, this work typically necessitates either complex
computations, heavy signalling/control overhead, and/or approximated
sub-optimal results. In this work, we employ and adapt the IEEE 802.11 protocol
in the NUM with a simple and efficient queue management mechanism. Unlike the
majority of the published work in this area, we focus on the feasibility of the
proposed solution in case of random static and mobile networks considering the
overheads and the signalling methods.
We propose a
novel algorithm that jointly solves the congestion, multipath routing, and
contention distributed control problem for MANETs. The objective is to find the
end-to-end optimal source rates at the transport layer, sub-flow rates for each
path of the multipath sessions at the network layer, and persistence probability
at the MAC layer. The primal problem formulation is a non-convex, non-separable
NUM optimization. By introducing new variables, applying certain
transformations, and using an analogy based on Ohm's law, we develop a
distributed algorithm that can find the optimal solution for general concave
utility functions.
The algorithms are implemented in NS-3 and evaluated against non-idealistic
scenarios, i.e. link failures, message losses, asynchronous updates, and with
the presence of inaccurate topology information. We evaluate the overhead and
signalling associated with the algorithms quantitatively and qualitatively and
provide absolute gain values. The results show that the proposed algorithms
significantly outperforms layered approaches, using standard protocols such as
TFRC.