Abstract:
In the last decade, wireless LANs (WLANs) based on the IEEE 802.11 standards have become
ubiquitous in our daily lives. During this time we have seen more than 10-fold increase in usage
and the number of wireless devices. To satisfy ever increasing demands, physical layer (PHY)
data rates in WLANs have scaled from a few Mbits/sec in 802.11g to hundreds of Mbits/sec in
IEEE 802.11n to over Gbits/sec in the IEEE 802.11ac standard. In addition, due to the emergence
of popular online services, such as YouTube and Netflix, there has been persistent traffic growth
due to real-time applications (e.g., video streaming). These trends bring about new performance
challenges that are likely to become problematic for high-speed WLANs: These challenges include
(a) achieving high user-level throughputs at high PHY data rates and (b) meeting the quality of
service requirements of diverse applications (e.g., video streaming, web surfing, and bulk transfers)
when they co-exist in a WiFi network.
Due to the shared nature of the wireless medium, a carrier sensing based random access protocol
is used in all 802.11-based standards. To arbitrate access to the channel, wireless access
protocols introduce overheads like backoffs, preambles, and acknowledgements that lower performance
efficiency at high data rates thereby resulting in low throughput. To address this inefficiency,
recentWiFi standards (e.g., 802.11n/ac) allow (a) frame aggregation, whereby multiple frames are
transmitted as a single aggregate frame on every channel access, and (b) block acknowledgements,
whereby a single frame is to used for acknowledging the receipt of several frames. These features
amortize the contention overhead over multiple frames and thus improve efficiency.
At high data rates, frame aggregation introduces two challenges. First, sending large aggregate
frames in a single transmission increases the opportunity cost of losing a frame, which leads to
greater degradation in performance. In WiFi networks, frame losses can occur due to a weak
signal, collisions, or hidden nodes. The MAC layer should respond differently to different types of losses. To achieve high performance, it is essential to infer the cause of frame loss accurately.
We propose, implement and evaluate BLMon, a framework for loss differentiation that uses loss
patterns within aggregate frames and their retries to achieve loss differentiation accurately and with
low overhead.
The second challenge arises in the presence of a mix of traffic, ranging from delay sensitive
real-time applications to bulk file transfers that require high throughput. We show that using QoS
mechanisms in high-speed WLANs presents a tradeoff between maximizing the performance of
real-time applications and achieving high throughput. We design SlickFi; a service differentiation
scheme that addresses this tradeoff and simultaneously maximizes the performance of real-time
applications and network throughput. SlickFi achieves this by (a) isolating different types of traffic
in non-overlapping parts of the spectrum by mapping them to different radios; and (b) adapting
channel width on a per-frame basis to make efficient use of the wireless channel.
The proposed solutions are readily deployable on commodity devices using only software level
changes. We demonstrate the validity of our solutions by performance evaluation over a real
testbed in diverse scenarios.