Intelligent Network Discovery for Next Generation Community Wireless Networks Andr´ es Arcia-Moret * , Antonio Araujo † , Jos´ e Aguilar ‡ , Laudin Molina § , Arjuna Sathiaseelan * * Computer Laboratory, University of Cambridge, Cambridge, UK † Centro Nacional de Desarrollo e Investigaci´ on en Tecnolog´ ıas Libres, M´ erida, Venezuela ‡ Universidad de Los Andes, M´ erida, Venezuela § Institut Telecom / Telecom Bretagne, Universit´ e Europ´ eene de Bretagne, Rennes, France Email: {andres.arcia, arjuna.sathiaseelan}@cl.cam.ac.uk, aaraujo@cenditel.gob.ve, aguilar@ula.ve, laudin@telecom-bretagne.eu Abstract—Nowadays, IEEE 802.11 networks are the most popular option to have wireless access to the Internet and a promising technology to tackle the digital divide that accounts for 2/3 of the world population. However, the popularity of these networks have raised a complex discovery and connection process, i.e., any device has to pass through an expensive scanning process of available Access Points in crowded and chaotic deployments. The scanning process can be modelled by a set of metrics exposing a trade-off between latency and the discovery rate when searching for appropriate Wi-Fi connection. Consequently, in order to improve the connection process, we use a multi-objective optimisation approach for generating optimal scanning sequences. We propose a framework to assist the network discovery within a Community Network, and we have adapted a Cultural Algorithm as an intelligent component for calculating optimal scanning sequences. Results show that we can derive optimal scanning sequences better than standard approaches for scanning in chaotic network deployments. I. I NTRODUCTION Today, IEEE 802.11 networks are the first option to have ubiquitous and low-cost wireless access to the Internet. We count on millions of Wi-Fi devices in many different new and challenging contexts: Community Networks [1], Sensor Networks [2], Internet of Things, and personal computers in Home Networks. In all of these scenarios, there has been a dramatic increase on the number of devices requiring to connect to Wi-Fi Access Points (AP). In order to connect to the wireless access network, a device must scan its surrounding and find an appropriate AP. However, the scanning process embedded in mobile and desktop devices does not follow a standard pattern or design principle. In Community Networks, people spontaneously deploy APs. Nomadic users are able to access thousand of community APs belonging to the same direct provider [3], and also have access to virtual network operators (VNOs) such as FON 1 . In both cases a user has to pass through a scanning process, being the most expensive sub-process within the discovery of a wireless topology. In this respect, the discovery process is becoming iterative and time consuming in Wi-Fi dense deployments. Recently, we have observed in [4] that in a regular discovery process, the device has to scan multiple times to discover available APs in a densely-covered urban area. World-wide trends show that this is likely to be a regular case 1 http://www.fon.com in the near future 2 . The current suboptimal behaviour of the scanning algo- rithm is present in the vast majority of devices, significantly consuming energy and impacting network performance [5]. The scanning traffic is becoming a potential problem lowering the available bandwidth and frequently interrupting regular transmissions. This is mostly because of the increasing number of devices using Wi-Fi and the congestion induced by non- adapted scanning process [4]–[6]. On the other hand, a Mobile Station (MS) may also perform a regular scanning by testing a partial or a complete set of channels. Approaches for partial scanning show that small bursts of scans allow uninterrupted execution of applications and look forward to perform seamless handovers. Whereas regular complete-set approaches for modern devices (iOS or Android based) perform sequential scanning without any adaptation on the sequence nor the duration of the timers. Considering that the later is most common scanning strategy [7], and that the nature of the scanning process represents about 80% of the handover time [8], an efficient scanning will not only represent an improvement on aggregated control traffic reduction for public Wi-Fi access [5], but also – if specific scanning sequences are given to specific group of users (§ III)– performing load-balancing in systems like VPuN [1] or Meraki 3 . A. Scanning process in 802.11 networks The scanning is the first (and most time consuming) sub- process for a client willing to attach to an IEEE 802.11 network, in which a client interface looks for available APs for later associate to them [9]. Although the ultimate (and common) goal of a scanning is to find all available APs to which the station might be able to join, it is very costly in terms of aggregated number of beacons and energy at the client. To discover all APs, i.e., the topology within a dense area, the device should properly be adjusted with a pair of timers, namely MinChannelTime and MaxChannelTime, referred as MinCT and MaxCT respectively from now on [10], [11]. IEEE 802.11 standard [9] defines two kinds of scanning 2 https://wigle.net/ 3 https://meraki.cisco.com/ ISBN 978-3-901882-80-7 © 2016 IFIP 2016 12th Annual Conference on Wireless On-demand Network Systems and Services (WONS) 81