With the exponential increase in wireless data traffic risking to overwhelm the current networking infrastructures, mmWave networking is seen as on of the most promising paths out of the logjam. IEEE introduced a new standard recently, 802.11ad, that uses mmWave frequencies for delivering multi-Gigabit capacities in a WLAN setting. Such capacities, however, do not come for free: mmWave frequencies have poor propagation characteristics and suffer from blockage, whose consequences on the networking protocols are not fully understood.In this paper, we aim at in-depth understanding of two key primitives in 802.11ad networks in environments rich with blockage: user association and reliability. Using careful measurements on COTS hardware, we find that user association under blockage takes an unreasonably long time, and, after analyzing the causes, provide practical insights on how to reduce this time by an order of magnitude. Relying on a multi-layer tracing tool allowing unprecedented granularity in the analysis of 802.11ad networks, we find that the part of the new standard regarding reliability is underspecified leading to different interpretations in practice. Their potential impact on performance is then evaluated using trace-based simulations, which reveal an interesting throughputreliability tradeoff. Finally, we find that TCP is poorly prepared for sudden changes in round-trip timeouts, as those induced by link blockage in mmWave networks, leading to an unnecessary increase in timeouts.

In the race for Gigabit wireless links, IEEE introduced 802.11ad, an amendment aimed at delivering Gbps capacities in a WLAN setting by leveraging the 60 GHz band. The key innovation of the standard is its beamforming training protocol. Executed periodically at the beginning of every beacon interval, it enables the formation of directional links. To address contention during the uplink part of beamforming training, 802.11ad introduced A-BFT (Association BeamForming Training), an Aloha-inspired, two-level backoff race. While central to the functionality of 802.11ad networks, the performance of A-BFT, however, remains poorly understood.In this paper, we propose an analytical finite-population model for evaluating the performance of IEEE 802.11ad A-BFT under two channel models: loss-free, and a channel introducing a constant bit error rate. After using an open-source simulator to demonstrate its accuracy, we use our model to assess the performance of A-BFT. We find that a counter-intuitive, quiteasily/be-lazy approach by the stations leads to the best overall beamforming training performance.