Understanding the role of active scans for their better utilization in large-scale wifi networks

Abstract

A WiFi client needs to discover access points (APs) in its vicinity before it can establish a WiFi connection and initiate a data transfer. Discovery is also essential for a client to stay connected to an AP deemed best at any given time. Discovery mechanisms in WiFi can be categorized into that of passive and active scanning. Passive scanning involves a client listening to beacon frames across WiFi channels to know which access points are in its vicinity. Using this method a client doesn’t create any additional load on the network. However, the latency in discovering new access points, especially in low-density AP deployments, can be significant and impacts the performance of latency-sensitive applications like VoIP. The alternative of active scanning involves a client proactively broadcast probe requests, which are responded to with a probe response by access points that receive it. The proactive sending of probes reduces latency in finding access points and is often used by clients. While simple and effective, we discover that this seemingly innocuous procedure of probing can cause severe degradation of throughput in an enterprise WiFi network, such as that of Cisco and Aruba. This motivates the premise of this thesis that management procedures in WiFi, which include discovery and association, need a thorough relook. They must be adapted to WiFi networks of today that have high densities of access points and clients unforeseen when the mechanisms were envisaged. Not only has the density of clients increased but also the heterogeneity of their service requirements. While some, for example streaming video, may desire high throughputs for extended periods of time, others, for example, temperature sensors, may only wake up intermittently to send small bursts of data. Last but not the least, mechanisms like probing are no longer used just for AP discovery but also support WiFi services like localization. Changes to mechanisms may require interventions that alleviate the impact on such services. We begin by investigating the protocol of active scanning from the perspeci tive of present-day large-scale WiFi networks. With several real-world network traces, we analyze the extent of the impact of active scanning. We demonstrate empirically that WiFi clients, typically in 2.4 GHz frequency band, often trigger active scans without a clear need for the same. As a consequence, they inject excessive low bit-rate traffic in the network that exponentially brings down the goodput of the network. We develop a metric that measures the growth of this traffic and an inference mechanism that detects the cause of this growth. We also suggest measures to mitigate these causes. While dealing with such unnecessary scans is desirable, we observe that a reduced frequency of active scans can have unintended consequences on services leveraging WiFi. Specifically, we study WiFi-based indoor localization that presumes that clients always scan. However, WiFi networks primarily operating in 5 GHz often experience a low rate of scanning and this severely deteriorates the performance of localization services. We demonstrate that an efficient floor detection mechanism can significantly reduce such localization errors. Lastly, we argue that for transferring small amounts of data such as infrequent sensor readings from IoT nodes, a WiFi client must not need to go through the entire process of connection establishment and maintenance. Here, we propose to leverage the perpetual WiFi active scans to enable such data transfer. The proposed approach of data transfer is not only beneficial in WiFi networks where establishing and maintaining a WiFi association is hard due to a variety of reasons, but also saves battery and reduces network traffic significantly. In our work, we will restrict our proposals to being device agnostic. Specifically, we don’t expect changes to be made at the device end. This allows for the possibility of their faster adoption in enterprise networks. Finally, we validate all our proposed solutions on real devices in live WiFi networks and demonstrate how these solutions enhance the performance of present-day WiFi networks.