Characterizing and Improving WiFi Latency
in Large-Scale Operational Networks
Kaixin Sui
†‡
, Mengyu Zhou
†
, Dapeng Liu
†‡
, Minghua Ma
†‡
,
Dan Pei
†‡∗
, Youjian Zhao
†‡
, Zimu Li
†
, Thomas Moscibroda
§
†
Tsinghua University
§
Microsoft Research
‡
Tsinghua National Laboratory for Information Science and Technology (TNList)
ABSTRACT
WiFi latency is a key factor impacting the user experience of
modern mobile applications, but it has not been well studied at large
scale. In this paper, we design and deploy WiFiSeer, a framework
to measure and characterize WiFi latency at large scale. WiFiSeer
comprises a systematic methodology for modeling the complex
relationships between WiFi latency and a diverse set of WiFi
performance metrics, device characteristics, and environmental
factors. WiFiSeer was deployed on Tsinghua campus to conduct a
WiFi latency measurement study of unprecedented scale with more
than 47,000 unique user devices. We observe that WiFi latency
follows a long tail distribution and the 90
th
(99
th
) percentile is
around 20 ms (250 ms). Furthermore, our measurement results
quantitatively confirm some anecdotal perceptions about impacting
factors and disapprove others. We deploy three practical solutions
for improving WiFi latency in Tsinghua, and the results show
significantly improved WiFi latencies. In particular, over 1,000
devices use our AP selection service based on a predictive WiFi
latency model for 2.5 months, and 72% of their latencies are
reduced by over half after they re-associate to the suggested APs.
Keywords
Wireless Network; WiFi Latency; Measurement
1. INTRODUCTION
WiFi latency is a critical performance measure for modern real-
time, interactive Internet applications, such as online gaming, web
browsing, and live collaboration applications. These applications
have strict requirements on end-to-end latency [28, 33, 53]. For
example, [53] shows that 20 ms last-mile latency will be amplified
into about two to three seconds of web page load time, so slow that
is beyond 47% of web users’ expectation [19]. While the latency
on the wired Internet is relatively stable and well-engineered (e.g.,
via CDN), the last-hop WiFi latency is often unpredictable and
a dominating factor in the end-to-end latency. For example, our
measurement results in §3.1 show that, while the median WiFi
∗
Dan Pei is the corresponding author.
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for components of this work owned by others than
ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from permissions@acm.org.
MobiSys’16, June 25-30, 2016, Singapore, Singapore
c
2016 ACM. ISBN 978-1-4503-4269-8/16/06. . .$15.00
DOI: http://dx.doi.org/10.1145/2906388.2906393
latency is as low as 3 ms, its 90
th
and 99
th
percentiles are around
20 ms and 250 ms, respectively.
Characterizing WiFi latency using WiFi-related metrics (e.g.,
channel utilization) is a key first step towards mitigating these
problems. Understanding the reasons for high WiFi latency oc-
currences can lead to better designs of wireless protocols/hardware
as well as better diagnosis and improvements in operational WiFi
networks. However, measuring and characterizing WiFi latency at
large scale is challenging, and to the best of our knowledge, there
exists no comprehensive attempt at this problem. The reasons are
manifold. First, measuring WiFi latency directly in an enterprise-
or campus-scale WiFi network is difficult because i) WiFi latency is
not reported in commonly used wireless products; ii) sniffer-based
measurement methods are too costly or inconvenient to deploy in
large-scale networks; iii) methods based on modifying firmware
or software of access points (APs) take significant time to be
adopted in commercial deployments; and iv) vanilla measurement
methods based on pings are influenced by the device’s energy
saving modes (§2.1). Secondly, while directly measuring WiFi
latency is thus challenging, characterizing WiFi latency by means
of reducing it to other more easily measurable WiFi-related metrics
is also challenging. Specifically, it turns out that there is no
simple correlation between WiFi latency and any other such readily
available metric. Instead, as we show in this paper, WiFi latency is
related to other radio parameters and metrics in subtle and non-
linear ways, revealing a complex interplay between multiple WiFi
metrics, protocol parameters and environmental factors.
To tackle these challenges, we design WiFiSeer (Fig. 1), a
framework for measuring and characterizing WiFi latency at large-
scale, as well as improving WiFi latency in large operational
Enterprise Wireless Local Area Networks (EWLAN) [9, 10], such
as in campuses, shopping malls, airports, and hotels.
WiFiSeer first makes use of a new ping based methodology,
called ping2 (§2.1), to measure the WiFi latency of devices
connected to the EWLAN. At the same time, WiFiSeer periodically
polls the readily available data using SNMP (Simple Network
Management Protocol) to derive potential impacting factors of
WiFi latency in the EWLAN. Data from ping2 and SNMP are
associated and form the WiFi latency dataset of WiFiSeer. In a
one-week exemplary dataset of real deployment in the EWLAN
of Tsinghua University (THU-WLAN), hundreds of millions of
WiFi latency records are collected based on ping2. Each of these
records associates a specific WiFi latency with 11 industry-standard
radio factors (e.g., channel utilization), 3 protocol factors (e.g.,
channel number), and 6 environmental factors (e.g., buildings) from
the SNMP data (§2.2). More than 2,700 APs, 114 buildings, and
more than 47,000 mobile devices are observed in the dataset.
剩余13页未读，继续阅读
资源评论
