Testbed
Dave Young & Bryan Cribbs
30 December 2004
On the growing CUWiN outdoor testbed, we are able to daily demonstrate the conditions that defeat pure shortest path, link-state routing for community networks. We are also discovering new properties of our radios and the radio environment, new community-networking possibilities.
The first important discovery is that even in the "urban jungle," extraordinarily long links are not as rare as we'd thought. Redundant links will be common, so the network will be very robust to individual node failures. Rainfall has very little direct effect on our network, however, we've observed that the rain-laden tree leaves have a powerful effect that outlasts the rainstorm. With greater node density and more redundant links, this will not be a serious problem. An important question to answer in the new year is, why do any packets pass through rain-laden tree leaves at all? It is possible that by intelligently adapting the link parameters (packet length and data rate), we can improve throughput during and after rain.
We have found out a subtle twist on the "IBSS splits" problem that we have described in this IBSS Splits Report and combatted previously. We did not think that Intersil Prism station firmware version 1.4.9 was susceptible to IBSS splits. Some early observations on our testbed suggest that in conditions where there is one-way communication, a defect in version 1.4.9 will cause splits. When the radio firmware stops receiving intelligible beacons from its neighbors, it enters a mode where it rotates through a random sequence of binary network identifiers, called BSSIDs. All stations' BSSIDs must match or else communication cannot take place.
The BSSID is broadcast in 802.11 beacons. After a station enters rotating-BSSID mode, its undetected neighbors may receive its beacons and change their BSSID to match its own. Communication between the neighbors and the neighbors' neighbors is momentarily cut off until the BSSIDs are synchronize. Thus a wave of disconnections moves outward from the station with the rotating BSSID. One-way communication is a fact of life in radio networks, and it is likely that the station in rotating-BSSID mode still does not detect its neighbors after its change of BSSID. It will change BSSID again. Another wave of disconnections occurs. This suggests that a denial of service (DoS) attack is possible; we will have to invent measures to protect against it. Banishing Intersil Prism radios from mesh networks is not a practical or sure-fire solution.
We are confident that capabilities of the latest crop of 802.11 radios,including the Atheros radios in our new testbed nodes, will let us protect against Prism flaws and DoS attacks.
Another discovery is that community wireless networks (CWNs) urgently need a protocol for selecting Internet gateways; default routes are not enough. All of our Internet-connected routers presently advertise a "default route" throughout the CWN. Any packet whose destination is not on the CWN is sent to the nearest gateeway through the default route. When default routes are advertised on the CWN, any node that is "equidistant" to two or more gateways is in danger of flip-flopping between the gateways. TCP/IP connections will ordinarily be severed every time a node flip-flops. That is CUWiN's principal concern, however, there are others.
All gateways are not created equal. In the first instance of flip-flopping that we observed on the Urbana testbed, the first gateway was a DSL, and the second was a 56K dialup! A smart gateway-selection algorithm would have chosen the DSL. CUWiN anticipates that in many CWN deployments, some node hosts will offer a for-pay Internet gateway, others will offer a free Internet gateway, and still others will impose capacity limits on their gateway.
