Powering Privacy: On the Energy Demand and Feasibility of Anonymity Networks on Smartphones

Many different anonymity networks have been designed and implemented over the last 20 years. These networks protect communication and metadata through multi-layered encryption and cover traffic. However, there is little research on whether such networks are actually practical on smartphones with limited battery power. This is important as these are the dominant computing devices today. Previous research suggests that cryptographic operations and background radio transmissions are the two main contributors to energy consumption when running such software on mobile devices. We develop and open-source a test setup that measures actual energy consumption, including side-effects that evade simple models. With this setup we explore the costs of cryptography, radio communication, and background scheduling. We find that radio communication dominates overall power consumption, while cryptographic operations (asymmetric and symmetric) are negligible for typical anonymity network workloads. We also investigate the feasibility of running anonymity networks continuously to protect the metadata of all communication. For a 12-hour usage period, a mobile Tor client on a 4G network requires an additional 4 percentage points of battery charge which appears practical and is at least as efficient as the commercial VPN clients that we tested. However, a mix network client that continuously sends cover traffic requires up to 30 percentage points for the same period. Such costs are unlikely to be acceptable to many users. Over the last few years end-to-end (E2E) encryption has reached mainstream popularity through mobile messaging applications such as Signal and WhatsApp with billions of users. E2E encryption ensures that only the intended recipients can read a message and that its content is hidden from anyone else, including the messaging service provider. This comes in addition to securing web-based communication through widely adopted standards such as HTTPS. However, as messages and packets are traveling through third-party infrastructure, most systems still leak metadata such as who is talking to whom, from where, and when. This metadata is easy to capture and interpret for messaging providers, network operators, and global adversaries. This can be dangerous for a whistleblower who reaches out to a journalist, an activist who is researching information online, and diplomats corresponding between embassies. These risks are even more pronounced when using smartphones. As these devices follow their owner throughout the day, they leak sensitive location data through changing IP addresses and other connection characteristics. Additionally, smartphones roam freely between many different networks such as WiFi and 4G. This exposes the user to many more infrastructure providers that they need to trust. The 2019 Hong Kong protests are one example that shows a large group organizing themselves using mobile devices and requiring access to secure, anonymous communication [1] . Anonymity networks protect metadata and we explore two points in the solution space covering real-time web-based and high-latency message-based communication. We include VPN clients in our evaluation as a baseline benchmark. Tor [15] is a popular deployment that disguises the real sender and recipient of web-based communication by routing it through multiple onion routers. Its encryption scheme ensures that each hop only learns about its direct predecessor and successor along the chosen path. This makes it hard for an adversary to follow messages through the system even if they have compromised some of the onion routers. However, a powerful adversary who suspects that two individuals in its influence area communicate with each other can perform traffic analysis to confirm their hypothesis. To do this the adversary captures the traffic patterns at both endpoints A and B. If A is indeed talking to B, an adversary would expect that messages sent by A result in messages received at B shortly afterwards. Other anonymity network designs, such as mix networks, offer strong metadata privacy for message-based communication to counter such traffic analysis. This is achieved through