The PhD: Cross Layer Performance Improvements in Multi-hop Ad-hoc Networks

I am a PhD student at [Murdoch University] in the school of IT. My PhD is in the area of wireless mesh networks, often known as multi hop ad hoc networks. I am approaching a number of different problems in this area with a cross layer perspective. This is an interesting new area and the solutions often require completely new and original perspectives. A basic, but important example of how multi hop ad hoc networks challenge traditional layered networking paradigms is routing. Routing is the biggest area of research within multi hop ad hoc networks and traditionally belongs at layer 3, or the network layer. However, a solution to the routing problem is being attempted by the IEEE with the 802.11s standardization effort ([TGs]). This group is implementing, among numerous other features, routing at the MAC layer. There is an open source group called [open802.11s] making this technology avaliable to everyone. Simultaneously there is a lot of IETF work with the [MANET] task force standardizing a range of routing protocols. Experimental FOSS work such as [OLSR], [BATMAN], [AODV] and [BABEL] have been implemented in real networks and in most cases at the IP layer using a flat (or MAC layer type) addressing structure. This is a facinating problem becuase it is mutually agreed that, a) packets should be routed between wireless nodes and b) the addressing structure should be flat. Thus whether you are designing a solution at the MAC layer of the network layer, you are violating traditional networking paradigms.

This is just a basic example of how this new area is being approached by numerous people from different layers. My topics of interest are routing protocols, channel selection, addressing, scalability and transport layer interactions and optimisations for multi hop ad hoc networks.

D-Proxy: Distributed Proxy

D-Proxy is a distributed proactive proxy that can provide reliability in unreliable wireless networks. The idea is that the current layer 2 ARQ (Automatic Repeat Request) mechanisms introduce a large overhead in many wireless technologies. Put simply, the acknowledgement mechanism used to provide reliability in ARQ consumes time that could be used for transmitting data and is therefore inefficient. D-Proxy is a TCP proxy that is capable of providing reliability to the link without wasteful positive acknowledgement mechanisms. D-Proxy uses sequence numbers to discover lost packets and will request packets to be resent. D-Proxy will reorder the TCP stream for in sequence delivery. D-Proxy is written in C and runs on Unix based operating systems. It provides excellent reliability to wireless links. When the ARQ acknowledgement mechanism is removed from WiFi, throughputs increase by 21%. Our paper can be found [here] and the code for D-Proxy can be found [here] .

Channel Selection with Multi Radio 802.11 Ad-hoc Networks

Despite over a decade of research in multi hop ad hoc networks, a fundamental performance limitation remains largely unsolved. Multi-hop throughput degradations occur when transmissions are hopped over multiple wireless nodes utilizing the same frequency. This limitation is ultimately a contention problem, however, it is exacerbated by a numerous other issues, specific to ad hoc networks. The solution to this problem, which has been attempted at many layers in ISO networking model, is to turn multi hop ad hoc networks from single channel networks into multi channel networks. RDCS (Routing Driven Channel Selection) is a channel selection mechanism that operates with the OLSR routing protocol. It circumvents multi-hop performance problems by enabling 802.11 mesh nodes equipped with multiple radios to intelligently utilize a range of frequencies, dramatically increasing performance.

We built our channel selection mechanism around the OLSR routing protocol. We tested this implementation with 8 dual radio PCEngines ALIX nodes running [Voyage] Linux. The code can be found [here] and the PhD chapter [here]. Alternatively, click on publications for our much shorter conference paper.