“Vendor-Affected” WLAN experimental results: A Pandora’s Box?

Experimental results are typically envisioned as the ultimate validation reference for any theoretical and/or simulation modelling assumptions. However, in the case of Wireless LANs, the situation is not nearly as straightforward as it might seem. In this paper, we discuss to what (large) extent measurement results may depend on proprietary undocumented algorithms implemented in the vendor-specific card/driver employed. Specifically, we focus on the experimental study of IEEE 802.11 b/g outdoor links based on the widely used Atheros/MADWiFi card/driver pair. We show that unexpected performance divergences do emerge in two classes of comparative experiments run on a same outdoor link: broadcast versus unicast, and 802.11 g versus 802.11 b. We provide experimental evidence that the severe performance degradation emerging in the broadcast case and in the 802.11 g case, are not caused by physical reasons, such as interference, multipath, or different PHY/MAC settings, such as preamble type/size, etc. Rather, they are ldquosimplyrdquo caused by two different Atheros/MADWiFi proprietary algorithms. The conclusions are quite worrying. On one side, the scientific community seems still largely unaware of the existence of such vendor-proprietary solutions. This is crucial as, in most cases, lack of awareness may mislead to erroneous interpretation of the results. On the other side, the inclusion of proprietary solutions in the equipments is a growing trend, and the two findings documented in this paper might be just the very first ones emerging out from a Pandorapsilas box...