Characterizing Smart Home IoT Traffic in the Wild

Abstract

As the smart home IoT ecosystem flourishes, it is imperative to gain a better understanding of the unique challenges it poses in terms of management, security, and privacy. Prior studies are limited because they examine smart home IoT devices in testbed environments or at a small scale. To address this gap, we present a measurement study of smart home IoT devices in the wild by instrumenting home gateways and passively collecting real-world network traffic logs from more than 200 homes across a large metropolitan area in the United States. We characterize smart home IoT traffic in terms of its volume, temporal patterns, and external endpoints along with focusing on certain security and privacy concerns. We first show that traffic characteristics reflect the functionality of smart home IoT devices such as smart TVs generating high volume traffic to content streaming services following diurnal patterns associated with human activity. While the smart home IoT ecosystem seems fragmented, our analysis reveals that it is mostly centralized due to its reliance on a few popular cloud and DNS services. Our findings also highlight several interesting security and privacy concerns in smart home IoT ecosystem such as the need to improve policy-based access control for IoT traffic, lack of use of application layer encryption, and prevalence of third-party advertising and tracking services. Our findings have important implications for future research on improving management, security, and privacy of the smart home IoT ecosystem.

Integrating the Manufacturer Usage Description Standard in The Modelling of Cyber-Physical Systems

“…Despite the high benefits of the MUD standard [9], one of the main limitations is its model, which is only capable to describe limited network layer security policies. Current literature has already taken note of this problem [29] [30] [9] and there are already some works proposing augmented behavioural profiles. In this sense, [27] defines a MUD profile generated from a security evaluation process including properties such as recommended key sizes or cryptographic primitives, [14] integrates a new module in the MUD model to include Medium-level Security Policy Language (MSPL) policies and whereas [31] extends the model by considering dynamic security aspects in the context of smart buildings.…”

Green Offloading in Fog-Assisted IoT Systems: An Online Perspective Integrating Learning and Control

“…Therefore, the average task arrival rate is 10 per time slot. For each new task, its size (in bits) is sampled randomly from the real-world distribution in [38]. The computation intensity of each task is set to be 1000 CPU cycles per bit.…”

Machine Learning Systems for Intelligent Services in the IoT: A Survey

“…Furthermore, high data traffic volumes can incur significant financial costs [147]. Devices that predominantly upload or download control traffic, for example home automation sensors, work appliances, health and wearable devices, account for a smaller portion of traffic volume [112] and are more likely to be affected by latency than bandwidth constraints. Applications that upload or download media content, like smart cameras, game consoles and smart TVs, place a greater burden on traffic volumes and are much more affected by bandwidth constraints.…”

“…Privacy leakage of user information pertaining to data values, location and usage compromises Manuscript submitted to ACM the anonymity of data providers [8]. Location-based services can infer device location based on communication patterns, while usage data can reveal sensitive temporal activity patterns [112]. For example, home occupancy and fine-grained appliance usage can be inferred from electric smart meter data [179].…”