Inférence de la charge dans les réseaux Wi-Fi : modèles et expérimentations
Nour El Houda Bouzouita · HAL (Le Centre pour la Communication Scientifique Directe) · 2022
As one of the cornerstones for delivering flexibility and ease of deployment, Wireless Local Area Networks (WLANs), especially IEEE 802.11, have been widely deployed in a variety of situations: homes, corporate or campus networks, public areas, which has led to the explosion of wireless data usage and the colossal rise of access points, smartphones, and various mobile devices. In such dense environments, a device seeking connectivity must choose among multiple available Wi-Fi networks that are within its radio range. However, the procedure for selecting an access point is still a striking concern and a critical ongoing challenge, especially in public areas (e.g., train stations, airports, malls, etc.), since it is based on simple criteria that do not relate to the quality of service that the device will experience. In particular, the network load is not taken into account even though it is a key parameter for the quality of service and experience. In this dissertation, we study the possibility/capacity for an unmodified vanilla device, especially a smartphone, to estimate the load of a network from local measurements in the user space with no interventions from the access points nor root permissions. The network load can be expressed in many ways. In this work, we consider the Busy Time Fraction (BTF), defined as the fraction of time the wireless medium is sensed busy due to successful or unsuccessful transmissions. In this regard, we propose relatively simple and versatile analytical Markovian models specific to the application of BTF estimation in the presence of the IEEE 802.11 frame aggregation scheme introduced in recent 802.11 amendments. We model and simulate different scenarios in which a device induces the UpLink (UL) or the DownLink (DL) mean aggregation levels, in the user space, of an aggregated deterministic probe traffic competing with the traffic present in the network that can aggregate or not its frames. We then propose a novel and practical method called Frame Aggregation based Method (FAM). It leverages the frame aggregation mechanism to estimate the network load through its BTF and characterize the network traffic type. FAM combines an active probing technique to measure the actual packet aggregation level and analytical Markov models that provide the expected rate as a function of the volume and nature of the traffic on the network. The performance evaluation of the proposed Markovian models and the method has been established with the aid of the ns-3 network simulator and experimental test-beds under several scenarios. Results have shown that our method FAM is able to infer the network load with a granularity based on six levels of network loads for the considered scenarios.