Non-pipelined relay improves throughput performance of wireless ad-hoc networks

A. Velayuthani, K. Sundaresan, R. Sivakumar
Proceedings IEEE 24th Annual Joint Conference of the IEEE Computer and Communications Societies.  
The communication model typically assumed in wireless ad-hoc networks is based on a traditional "pipelined relay" (PR) strategy. In PR, an end-to-end flow has multiple outstanding packets (or data units) along the path from the source to the destination. In this paper, we argue that due to several unique properties of wireless ad-hoc networks, PR can be fundamentally improved upon. We present a new non-pipelined relay (nPR) strategy, where end-to-end flows have exactly one outstanding packet
more » ... data unit) along the end-to-end path. We show that nPR has the following properties: (i) Under idealized network conditions, it provides performance improvement, in terms of end-to-end throughput capacity and network transport capacity over PR, and achieves proportional fairness; and (ii) Under practical network conditions, it further increases the above performance improvements, both in terms of the throughputs achieved, and in terms of the fairness between flows. Finally, we present a forwarding protocol that practically realizes nPR. Through analysis and ns2 based packet level simulations, we evaluate the performance of the proposed strategy, and that of the forwarding protocol. ¢ ¡ © ¦ . Over the last few years, several approaches have been examined to transcend the capacity limits established in [1] . In [2], an approach is presented to leverage multi-user diversity, in the presence of node mobility, to achieve ¢ ¡ ¥ § ¦ network transport capacity. However, the approach is appropriate only for delay insensitive applications, where the "insensitivity" requirements can be from a few seconds to a few hours [2]. In [3], [4], the results in [1] are extended for wireless ad-hoc networks with smart antennas. The conclusions are that only an ¡ ! " ¡ ¥ § ¦ # ¦ improvement in throughput can be achieved even if arbitrarily complex signal processing in terms of beam forming capabilities is assumed. Finally, in [5] , the results in [1] and [2] are extended through the derivation of delay bounds, and identification of optimal delaythroughput trade-offs. In this paper, we revisit the network model considered in [1]. However, we show that, through a simple change in the network communication strategy, fundamental improvements in performance can be achieved for the end-to-end throughput capacity and network transport capacity, in an idealized network setting (with optimal centrally coordinated medium-access control, routing, and relaying). Perhaps, equally importantly, we show that the change also delivers significant additional performance ben-efits in a practical wireless ad-hoc environment using distributed protocols for medium-access control, routing and packet relaying. Essentially, we term the communication strategy modeled in [1] as a pipelined relay (PR) strategy, where several packets belonging to an end-to-end flow simultaneously wait to be served at different stages along the flow's path. Note that such a strategy is the default assumed in most, if not all, related works on both the theory and practice of wireless ad-hoc networks. In this context, we examine an alternate non-pipelined relay (nPR) communication strategy, where every end-to-end flow in the network, by default, has exactly one outstanding packet along its path. We establish that this simple change in the communication strategy can result in an ¢ ¡ $ ¡ ¥ § ¦ # ¦ performance improvement in the end-to-end throughput capacity in an idealized network setting. In addition, under the same idealized setting, the strategy results in proportional fair allocation of network resources, and improves the network transport capacity. Note that the latter improvement does not necessarily follow from the earlier mentioned increase in end-to-end transport capacity, and is an additive improvement. Equally importantly, we demonstrate that nPR (i) reduces the degree of contention in the network thereby considerably improving both the utilization and fairness properties of practical distributed medium access control (MAC) schemes such as CSMA and its variants, (ii) ameliorates the impact of route failures on flows by virtue of the shorter time flows need to be active given the improved throughput capacity, and (iii) provides temporal decoupling between network flows that enables effective load balanced routing to be performed, unlike in PR where load balanced routing has been shown to be ineffective due to the high degree of coupling between flows. Finally, we present a distributed forwarding protocol (DFP) that addresses the unique challenges required to realize the nPR strategy practically in a wireless ad-hoc network. Thus, the contributions of this paper are threefold: % In an idealized network setting, we show that nPR can achieve fundamental improvements in end-to-end throughput capacity, and network transport capacity, while achieving proportional fairness. % We show that nPR brings in additional performance benefits in a practical wireless ad-hoc network environment using distributed protocols for medium access control, and routing. % Finally, we present the DFP distributed forwarding protocol for wireless ad-hoc networks that realizes the nPR strategy, and tackles the unique challenges that arise in the process. The rest of the paper is organized as follows: In Section II, we present terminology and definitions for the models considered in the paper. In Section III, we describe the performance improve-ments achievable when using nPR and we also present quantitative results to demonstrate the improvements due to nPR under practical conditions. In Section IV, we present the distributed forwarding protocol that realizes nPR. In Section V, we study the performance of DFP. We discuss the issues in Section VI. Finally, in Section VII we present the related work, and present conclusions in Section VIII. II. DEFINITIONS AND MODELS In this section, we outline the different paradigms of communication that are of interest in this work. However, to make the discussions convenient we first present certain basic definitions that shall be used in the rest of the paper. A. Definitions Network: A wireless ad-hoc network consisting of
doi:10.1109/infcom.2005.1497916 dblp:conf/infocom/VelayuthamSS05 fatcat:sn5vu24jyzge3iq6jlprmgutm4