LAP: Lightweight Anonymity and Privacy
Total Page:16
File Type:pdf, Size:1020Kb
2012 IEEE Symposium on Security and Privacy LAP: Lightweight Anonymity and Privacy Hsu-Chun Hsiao† Tiffany Hyun-Jin Kim† Adrian Perrig† Akira Yamada‡ Samuel C. Nelson§ Marco Gruteser§ Wei Meng †CyLab/CMU ‡KDDI Labs §Rutgers University Tsinghua University Abstract—Popular anonymous communication systems often Attack class require sending packets through a sequence of relays on dilated No anonymity protection paths for strong anonymity protection. As a result, increased Lightweight anonymity end-to-end latency renders such systems inadequate for the Low-latency anonymity majority of Internet users who seek an intermediate level of End-server LAP High-latency anonymity anonymity protection while using latency-sensitive applications, such as Web applications. This paper serves to bridge the gap Government between communication systems that provide strong anonymity class Tor [1] protection but with intolerable latency and non-anonymous Global communication systems by considering a new design space eavesdropper Optimal solution Mix network [8] for the setting. More specifically, we explore how to achieve near-optimal latency while achieving an intermediate level of Latency anonymity with a weaker yet practical adversary model (i.e., Figure 1. The design space of anonymous schemes. protecting an end-host’s identity and location from servers) such that users can choose between the level of anonymity may be impatient to wait.2 and usability. We propose Lightweight Anonymity and Privacy (LAP), an efficient network-based solution featuring lightweight Despite existing work that attempts to protect end-users’ path establishment and stateless communication, by concealing anonymity [1], [7], [8], it still remains a challenge to provide an end-host’s topological location to enhance anonymity against an intermediate level of anonymity and privacy protection remote tracking. To show practicality, we demonstrate that without introducing much latency. In this paper, our main LAP can work on top of the current Internet and proposed goal is to bridge the chasm between systems that provide future Internet architectures. strong anonymity with high latency and systems that support no anonymity with zero latency, and explore how to support lightweight anonymity and privacy that is efficient enough I. INTRODUCTION to protect all traffic. Note that those end-users who want an intermediate level of privacy primarily desire to remain Staying anonymous in today’s Internet requires anony- anonymous from servers such that servers cannot track mous overlay systems, such as Tor [1], to conceal the their behavior. This implies that guaranteeing the end-user’s communicating endpoint’s IP address, as it can reveal the anonymity and privacy against a single remote entity rather end-user’s identity and location [2]. Such overlay systems than a strong, global attacker may be a suitable relaxation attempt to facilitate anonymous communication using layer- of the attacker model to gain higher efficiency. encrypted packets traveling through indirect routes. How- We propose a new setting that we call Lightweight ever, this results in additional latency due to long end-to-end Anonymity and Privacy (LAP-setting for short) for private path length and cryptographic operations indirectly traveling and anonymous communication in the Internet with the through three Tor relays would be approximately four times following properties: slower than traveling along a non-dilated path. Moreover, Tor relays are constantly overloaded [3], further worsening • Low-stretch anonymity: packets for anonymous and the latency and throughput. Measurements show that the private communication should travel through near- average time to fetch an HTTP header using Tor is 4.04s — optimal routes such that the increase in the number ten times higher than fetching it without Tor [4]. Although of Autonomous Domains (ADs) normalized over the privacy-anxious users may tolerate seconds of latency for original path length is low. strong privacy, users desiring an intermediate level of privacy • Relaxed attacker model: an intermediate level of for default protection of daily online activities (e.g., prevent privacy can be achieved with sender and receiver websites from tracking them for behavioral advertising1) anonymity and location privacy. Hence, we relax the strong attacker model (e.g., global or government-class 1Users might enable the DO-NOT-TRACK option supported by most 2Studies have shown that online users are sensitive to waiting time: mainstream browsers. However, a recent study [5] has shown that this Amazon’s sales dropped by 1% for every 100ms increase in page load mechanism is hard to use due to configuration complexity and provides time, and Google’s ad revenue decreased by 20% for a 500ms increase in no guarantee as it depends on the self-regulation of online organizations. search result display time [6]. © 2012, Hsu-Chun Hsiao. Under license to IEEE. 506 DOI 10.1109/SP.2012.37 attackers) considered by existing anonymity systems. mechanisms for other layers, as recognized by previous As Figure 1 shows, our aim is to address a relaxed attacker network-based proposals [7], [14]. model (e.g., end-server attack) with near-optimal latency Contributions. while existing work addresses stronger attacker models 1) We explore the design space of anonymous protocols (e.g., government class or global eavesdropper) with higher in the context of a relaxed adversary model. latency. Although low-latency designs are shown to be 2) We propose Lightweight Anonymity and Privacy inherently vulnerable to a global eavesdropper, some users (LAP), an efficient network-based solution that en- who trust their local ISPs can achieve much higher efficiency ables lightweight path establishment and efficient for- under the LAP-setting. warding. Our mechanism, Lightweight Anonymity and Privacy 3) We evaluate LAP’s security and performance advan- (LAP), is an efficient and practical network-based solution tages. Our systematic analysis and the evaluation of featuring lightweight path establishment and efficient com- our software implementation confirm that LAP can munication. LAP attempts to enhance anonymity by obscur- improve anonymity with low performance overhead. ing an end-host’s topological location, based on two building blocks: packet-carried forwarding state, and forwarding-state II. PROBLEM DEFINITION encryption. We study how to camouflage an end-host’s topological lo- • Packet-carried forwarding state: each packet carries cation (i.e., potential origin within a given topological neigh- its own forwarding state such that ADs can determine borhood) in a network architecture to enhance anonymity the next hop from the packet without keeping local per- and location privacy in a practical manner. More specifically, flow state. we study how to design an anonymous forwarding protocol • Forwarding-state encryption: existing anonymity sys- that can protect the identities and locations of end-hosts from tems require entire packets to be decrypted/encrypted a weaker yet practical adversary, while demanding minimal as they travel using shared keys between the sender increase in latency. We do not claim to achieve complete and intermediate relays. In contrast, LAP allows each anonymity, but rather focus on providing an intermediate AD to use a secret key (known to the AD only) level of anonymity. to encrypt/decrypt forwarding information in packet In this section, we scope our problem in terms of desired headers. As a result, an AD’s forwarding information properties, assumptions, and threat model. can be hidden from all other entities while a LAP packet remains the same at each hop. A. Desired Privacy Properties LAP is extremely lightweight in the sense that (i) it Sender/receiver anonymity. Anonymity can be viewed introduces minimal overhead over non-anonymous packets as being unidentifiable within a set of subjects (e.g., users), in terms of latency and computational overhead on routers, also known as an anonymity set [15]. This implies that a (ii) it does not require any per-flow state to be stored on sender or a receiver can achieve stronger anonymity if its routers, and (iii) no separate keys are required to be set identity is hidden in a larger anonymity set [16].3 As a up with routers. In addition to its performance advantages, result, an attacker cannot link the sender and receiver if LAP’s unique design provides two additional merits. First, either sender anonymity or receiver anonymity is achieved. LAP supports different privacy levels such that an end-host Since the design of a full anonymous communication system can trade privacy for improved performance. Second, LAP is that can defend against timing attacks and conceal unique a generic design that can work with a wide range of routing platform characteristics is beyond the scope of this paper, protocols, which includes the inter-domain routing protocol we focus on concealing an end-host’s network identifier and BGP and new proposals such as SCION [9] and Mobility- location in the network topology (which we call “topological First [10]. Furthermore, we show that LAP fits especially anonymity”), which is an important step towards improving well with proposed routing protocols that support packet- sender/receiver anonymity. For simplicity in expression, carried forwarding state, such as SCION and ICING [11]. we also abbreviate “topological anonymity” simply with In this paper we focus on network-based solutions, where “anonymity” in the paper. users and locations can be identified