Title: The Feasibility And Usefulness Of Link Abstraction In Wireless Networks
Abstract:
Most recent measurement work in wireless networks has found that the concept of link abstraction does not hold for a wireless link. Research work has thus focussed on working around the absence of a link abstraction by doing intelligent things at the higher layers. Our measurement work in this thesis, has focussed on studying the feasibility and usefulness of a link abstraction especially in the two domains of wireless sensor networks and WiFi-based wireless mesh networks. We used Tmote Sky motes equipped with 2.4 GHz radios for our wireless sensor network measurements. The WiFi-based wireless measurements were carried out using 802.11b based Senao PCMCIA cards fitted in laptops and connected to external antennas.
Our experimental data suggests that in the absence of external interference, it is indeed possible, to achieve link abstraction in case of a wireless link. We find that there exists a threshold value of RSSI above which the error rates on the link are stable and negligible i.e. the wireless link becomes a close approximation of a wired link - a situation in which the link abstraction holds.
A link operating close to the RSSI threshold experiences variable error rates. This variability in error rate means that routing metrics proposed for Wireless Mesh Networks (WMNs), like ETX that rely on the Packet Success Rate will be inherently unstable. Imposing an RSSI threshold also implies that the maximum link range reduces, effecting a trade-off between link range and predictable performance. However, there are ways to increase the link range. We explore one of them - use of external antennas. The results, from our experiments carried out with wireless sensor nodes suggest that the use of external antennas can substantially increase the maximum communication range over which predictable performance is achievable up to 500m in case of a clear Line-of-Sight (LoS) being available. This range is more than sufficient for a wide range of sensor network applications. In addition, in some cases, it may also help to reduce the network to a single-hop network thereby simplifying the design of routing protocols and reducing the increased losses during multi-hop data transfer.
Finally, the feasibility of link abstraction, has a number of implications. It can help to plan predictable links with low, stable error rates in new networks and also help to decide the transmit powers to achieve stable, low error links in existing networks. Link abstraction can also be used to classify existing links in a network. By classifying an existing link with a stable, low error rate as 'up' and one with an intermediate error rate as 'down', we can restrict a network to use only those links with stable, low error rates so that predictable performance can be achieved. In wireless sensor networks, link abstraction can also simplify routing, which may eventually help to reduce the overhead of routing messages and thus increase the longevity of the network.