UCLA

As the Wi-Fi technology becomes pervasive in reality, its usage pattern is also turning highly diversified in operation settings and application scenarios. This consequently leads to new design requirements for Wi-Fi networking solutions in terms of various combinations in energy efficiency and latency, in addition to the throughput. However, the state-of-the-art solutions typically chase for high speed in the different generations of Wi-Fi technologies (from 802.11a/g to 802.11n/ac) at the cost of other metrics. For example, Multiple-Input Multiple-Output (MIMO) is widely used to boost speed, yet it consumes much more power. Higher throughput may yield long-tail loss behaviors and compromise the perceived latency for data transfer. Fundamentally, we believe that both the user demand pull and the technology push call for customizable Wi-Fi solutions.

In this dissertation, we describe our effort on customizable Wi-Fi technology. We propose solutions that seek to meet the diverse requirements (i.e., energy, throughput, and latency). Specifically, we have come up with technical results on three topics. The first one explores to use rate adaptation (RA), the mechanism critical to performance yet unspecified by the 802.11 standards, for energy efficiency. It is shown that current MIMO RA algorithms in 802.11n/ac are not energy efficient despite ensuring high throughput. Marginal throughput gain is achieved at much higher energy cost. We then propose EERA, an energy-efficient RA solution. It balances throughput for energy savings while meeting the data rate quest by applications. In the second topic, we target home Wi-Fi scenario and interactive gaming applications. We examine the millisecond-level latency requirements of such applications. We show that current solutions work well for throughput but not for latency, due to the long tail of the packet delay distribution. We thus propose LLRA, a new latency-aware RA scheme that reduces the tail latency. It takes concerted design in rate control, frame aggregation scheduling and software/hardware retransmission dispatching. The implementation and evaluation confirm the viability of both EERA and LLRA. In the third topic, we propose HetRA, which can meet heterogeneous goals at each client. It explores a new theoretical approach, called piecewise linear wireless service curve. Using this new solution framework, we can meet minimum throughput on a per-flow basis, while simultaneously improving energy efficiency at the device level. We show in both analysis and experiments that, the resulting design outperforms all existing RA algorithms that only pursue a single goal. In all three concrete designs, we confirm that, the upcoming Wi-Fi technology has to be customized to the given usage scenario and designated goals, and this involves complex tradeoffs along multiple performance metrics of throughput, latency and energy saving and along various granularities of data flows, flow aggregates, and devices within a single user client and among multiple users.