A Survey on Routing in Anonymous Communication Protocols

TLDR: In this article, the authors survey previous research on designing, developing, and deploying systems for anonymous communication and provide important insights about the differences between the existing classes of anonymous communication protocols.

Abstract: The Internet has undergone dramatic changes in the past 2 decades and now forms a global communication platform that billions of users rely on for their daily activities. While this transformation has brought tremendous benefits to society, it has also created new threats to online privacy, such as omnipotent governmental surveillance. As a result, public interest in systems for anonymous communication has drastically increased. In this work, we survey previous research on designing, developing, and deploying systems for anonymous communication. Our taxonomy and comparative assessment provide important insights about the differences between the existing classes of anonymous communication protocols.

Figures

Table 2. Routing Classification of Anonymous Communication Protocols: Mixnet Protocols

Table 5. Routing Classification of Anonymous Communication Protocols: DCnets and Miscellaneous Protocols

Fig. 3. The basic concept of Crowds: In this simple scenario, two users want to send their messages (message A and message B) to Website A and Website B, respectively. First, they send their message to a randomly selected node (step a1) and (step b1). The receiving node flips a coin to decide whether to send the message to another node or to its final destination. The receiving node of the message A randomly decides to send the message to another node (step a2), while the receiving node of the message B decides to send the message to its final destination (step b2). The receiving node of message A, again by flipping a coin, decides to send the message to another node (step a3). The message goes further until a node decides to send it to the final destination (step a4). Finally, the third receiving node decides to forward the message to its destination (step a4).

Fig. 4. An overview of a general DHT network (left image) and a DHT network where the search algorithm for routing indexes uses binary trees (image on the right), such as Salsa or DHT networks using the Kademlia look-up method.

Table 6. Overview of the Adversary Assumptions, Focus of Routing Feature, and Challenges That Our Four Routing Classes Face

Table 1. Overview of the Protocol Routing Characteristics

Full text

51
A Survey on Routing in Anonymous
Communication Protocols
FATEMEH SHIRAZI, KU Leuven, ESAT/COSIC and imec
MILIVOJ SIMEONOVSKI, CISPA, Saarland University, Saarland Informatics Campus
MUHAMMAD RIZWAN ASGHAR, Cyber Security Foundry, The University of Auckland
MICHAEL BACKES, CISPA Helmholtz Center i.G., Saarland Informatics Campus
CLAUDIA DIAZ, KU Leuven, ESAT/COSIC and imec
The Internet has undergone dramatic changes in the past 2 decades and now forms a global communication
platform that billions of users rely on for their daily activities. While this transformation has brought tremen-
dous benets to society, it has also created new threats to online privacy, such as omnipotent governmental
surveillance. As a result, public interest in systems for anonymous communication has drastically increased.
In this work, we survey previous research on designing, developing, and deploying systems for anonymous
communication. Our taxonomy and comparative assessment provide important insights about the dierences
between the existing classes of anonymous communication protocols.
CCS Concepts: • Security and privacy → Pseudonymity, anonymity and untraceability;
Additional Key Words and Phrases: Anonymous communication, routing protocols, communication networks
ACM Reference format:
Fatemeh Shirazi, Milivoj Simeonovski, Muhammad Rizwan Asghar, Michael Backes, and Claudia Diaz. 2018.
A Survey on Routing in Anonymous Communication Protocols. ACM Comput. Surv. 51, 3, Article 51 (June
2018), 39 pages.
https://doi.org/10.1145/3182658
1 INTRODUCTION
The Internet has evolved from a mere communication network used by millions of users to a
global platform for social networking, communication, education, entertainment, trade, and po-
litical activism used by billions of users. In addition to the indisputable societal benets of this
transformation, the mass reach of the Internet has created new powerful threats to online privacy.
This work was supported by the Research Council KU Leuven C16/15/058 and through KU Leuven BOF OT/13/070. In
addition, it was supported by the European Commission project H2020-DS-2014-653497 PANORAMIX and by Microsoft
Research through its Ph.D. Scholarship Programme. Moreover, this work was supported by the German Federal Ministry
of Education and Research (BMBF) through funding for the Center for IT-Security, Privacy and Accountability (CISPA)
(FKZ: 16KIS0345) and by the European Research Council Synergy Grant imPACT (n. 610150).
Authors’ addresses: F. Shirazi and C. Diaz, KU Leuven ESAT/COSIC, Kasteelpark Arenberg 10 - Bus 2452, B-3001 Leuven-
Heverlee, Belgium; emails: {fatemeh.shirazi, claudia.diaz}@esat.kuleuven.be; M. Simeonovski and M. Backes, CISPA, Cam-
pus E 9-1, 66123 Saarbruecken, Germany; emails: {milivoj.simeonovski, backes}@cispa.saarland; M. R. Asghar, Building
303S - Room 585, Department of Computer Science, The University of Auckland, Private Bag 92019, Auckland 1142, New
Zealand; email: r.asghar@auckland.ac.nz.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee
provided that copies are not made or distributed for prot or commercial advantage and that copies bear this notice and
the full citation on the rst page. Copyrights for components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires
prior specic permission and/or a fee. Request permissions from permissions@acm.org.
© 2018 ACM 0360-0300/2018/06-ART51 $15.00
https://doi.org/10.1145/3182658
ACM Computing Surveys, Vol. 51, No. 3, Article 51. Publication date: June 2018.

51:2 F. Shirazi et al.
The widespread dissemination of personal information that we witness today in social media
platforms and applications is certainly a source of concern. The disclosure of potentially sensitive
data, however, not only happens when people deliberately post content online but also becomes
possible inadvertently by merely engaging in any sort of online activities. This inadvertent data
disclosure is particularly worrisome because non-expert end-users cannot be expected to under-
stand the dimensions of the collection taking place and its corresponding privacy implications.
Widely deployed communication protocols only protect, if at all, the content of conversations,
but do not conceal from network observers who is communicating with whom, when, from where,
and for how long. Network eavesdroppers can silently monitor users’ online behavior and build up
comprehensive proles based on the aggregation of user communications’ metadata. Today, users
are constantly tracked, monitored, and proled, both with the intent of monetizing their personal
information through targeted advertisements and by nearly omnipotent governmental agencies
that rely on the mass collection of metadata for conducting dragnet surveillance at a planetary
scale.
Anonymous Communication (AC) systems have been proposed as a technical countermeasure
to conceal from network observers who is communicating with whom, when, from where, and
for how long, mitigating the threats of communications surveillance. The concept of AC systems
was introduced by Chaum (1981) in 1981, with his proposal for implementing an anonymous email
service that aimed at concealing who sent emails to whom. The further development of this concept
in the past 2 decades has seen it applied to a variety of problems and scenarios, such as anonymous
voting (Sako and Kilian 1995; Jakobsson et al. 2002), Private Information Retrieval (PIR) (Dingledine
et al. 2000), censorship-resistance (Waldman et al. 2000; Waldman and Mazières 2001), anonymous
web browsing (Goldschlag et al. 1996), hidden web services (Dingledine et al. 2004), and many
others.
Public interest in AC systems has strikingly increased in the past few years. This could be ex-
plained as a response to recently revealed dragnet surveillance programs, the fact that deployed
AC networks seem to become (according to leaked documents
1
) a major hurdle for communica-
tions surveillance, and to somewhat increased public awareness on the threats to privacy posed
by modern information and communication technologies.
The literature oers a broad variety of proposals for anonymity network designs. Several of
these designs have been implemented, and some are successfully deployed in the wild. Of the
deployed systems, the most successful example to date is the Tor network, which is used daily by
about two million users (The Tor Project 2017).
Existing designs take a variety of approaches to anonymous routing for implementing the AC
network. Routing determines how data is sent through the network, and it as such is the core
element of the AC design, determining to a large extent both security and performance of the
system. State-of-the-art approaches rely on dierent threat models and sets of assumptions, and
they provide dierent guarantees to their users. Even though survey articles on AC systems ex-
ist (Erdin et al. 2015; Sampigethaya and Poovendran 2006; Conrad and Shirazi 2014; AlSabah and
Goldberg 2015; Ren and Wu 2010; Edman and Yener 2009; Danezis and Díaz 2008; Serjantov 2004;
Raymond 2000), we still lack a systematic understanding, classication, and comparison of the
routing characteristics of the plurality of existing AC approaches.
The purpose of this survey is to provide a detailed overview of the routing characteristics of
current AC systems and to discuss how their routing features impact anonymity against dierent
types of adversaries, as well as overall performance. To this end, we rst identify the routing char-
acteristics that are relevant for AC protocols and provide a taxonomy for clustering the systems
1
https://wikileaks.org/.
ACM Computing Surveys, Vol. 51, No. 3, Article 51. Publication date: June 2018.

A Survey on Routing in Anonymous Communication Protocols 51:3
with respect to their routing characteristics, deployability, and performance. Then, we apply the
taxonomy to the extensive literature on AC systems, in particular including Mixnets, DCnets, Tor-
related systems, and Random Walk/Distributed Hash Table (DHT)-based protocols. To select AC
protocols for our examination we chose systems that have presentation value in terms of routing.
Most of the reviewed protocols are systems designed as overlay networks. We excluded next-
generation Internet AC solutions such as Sankey and Wright (2014) and Hsiao et al. (2012). Finally,
we discuss the relationship between dierent routing decisions, and how they aect performance
and scalability.
Outline. Section 2 provides our taxonomy for anonymous routing and describes the various
routing features and dimensions that we are considering for our evaluation and discusses the rela-
tionship between these routing features. Section 3 gives a compact tabular overview describing the
classication of existing systems in our taxonomy and reviews existing AC systems with respect
to their routing characteristics. Section 4 compares the four main categories of AC protocols in
terms of anonymity goals against dierent types of adversaries, scalability, and their applications.
Section 5 concludes the paper.
2 ANONYMOUS ROUTING PROTOCOL CHARACTERISTICS
In this section, we introduce the routing characteristics, deployability, and performance metrics
considered in our taxonomy, and we discuss the relationship between these characteristics.
2.1 Routing Characteristics
Generally, routing in a communication network refers to the selection of nodes for relaying com-
munication through the network. Routing schemes, however, require some essential design com-
ponents. For anonymous communication, we consider four building blocks that are relevant to
routing in AC networks. These building blocks are node management, transfer/retrieval of node
information to/by the routing decision maker, path selection, and forwarding or relaying; where
path selection is the main design component of routing schemes for AC protocols.
Several taxonomies and classications for routing protocols have been proposed in the litera-
ture (Bell and Jabbour 1986; Feeney 1999; Zou et al. 2002). However, AC networks aim to conceal
the metadata of communications and thus have security requirements that make them fundamen-
tally dierent from other networks.
In this section, we present a classication for anonymous routing protocols (see Table 1). Our
classication (see Tables 2 and 5) is an adaptation of Feeney’s taxonomy (Feeney 1999), which
classies the routing characteristics of mobile ad hoc networks into four categories:
(1) Communication model describes whether the communication is based on a single- or multi-
channel.
(2) Structure describes whether or not nodes are treated equally.
(3) State information describes where the topology information is maintained.
(4) Scheduling describes whether the information about routes is maintained at the source or
is instead computed on-demand.
This taxonomy does not address several relevant design features of AC networks, such as prob-
abilistic node selection for constructing circuits and security considerations for protecting routing
information from dierent network adversaries. In addition, not all the characteristics identied by
Feeney are relevant to AC routing. For example, the distinction between single- and multi-channel
features is not relevant to overlay networks, which constitutes a standard design choice for many
AC networks.
ACM Computing Surveys, Vol. 51, No. 3, Article 51. Publication date: June 2018.

51:4 F. Shirazi et al.
Table 1. Overview of the Protocol Routing Characteristics
We redene Feeney’s criteria to account for design choices that are relevant to anonymous
routing protocols. Nevertheless, we distinguish three groups of features inspired by Feeney’s cat-
egories: network structure, routing information,andcommunication model:
(1) Network structure describes the characteristics of the anonymous relays, the connections
between them, and the underlying network topology.
(2) Routing information describes the network information available to entities deciding on
the route of an anonymous connection.
(3) Communication model denes the entities that make the routing decisions and describes
how these decisions are made.
In what follows, we describe these features in more detail, including their various sub-features
and corresponding notation symbols used to denote individual feature instantiations. We refer to
Table 1 for a general overview of the resulting taxonomy.
2.1.1 Network Structure. We consider rst the network features that are relevant to anony-
mous routing. These are, specically, features related to: (a) the topology of the network, which
describes how nodes are connected; (b) the connection type, describing the characteristics of the
connections between nodes; and (c) symmetry, describing whether the entities participating in
the network are all similar, or if they can take on dierent roles and responsibilities for routing
data through the network.
(a) Topology. The topology describes the arrangement of various elements of the network,
such as routers and communication links between those routers. We only take the logical
topology of the network into account, which determines how data ows within it. We
note that physical topology characteristics, such as the geographical location of comput-
ers, sometimes matters in anonymous routing decisions, for example, when considering
ACM Computing Surveys, Vol. 51, No. 3, Article 51. Publication date: June 2018.

A Survey on Routing in Anonymous Communication Protocols 51:5
adversaries who control an Autonomous System (AS) (Feamster and Dingledine 2004;
Edman and Syverson 2009).
We consider the network as a graph in which the routers are represented by graph
nodes. An edge between two nodes exists if the routing strategy allows both nodes to be
directly connected as part of the same anonymous circuit.
The connectivity of nodes varies widely across AC network designs, and the advantages
and disadvantages of high and low levels of connectivity have been the subject of debate
for over a decade (Böhme et al. 2005).
Restricted routing proposals (Danezis 2003a) have shown that for applications that
are latency-tolerant, partially connected networks with certain topological characteris-
tics (e.g., based on expander graphs) provide optimal anonymity and latency trade-os
and mitigate certain attacks. These results further emphasize the impact of network con-
nectivity features for anonymous routing.
We classify anonymity networks into three categories according to their connectivity:
fully connected, mostly connected,andpartially connected networks.
• We consider a network to be fully connected ()
2
when nodes can potentially connect to
most(orall)othernodes.Notethatourruleofthumbisthatanodeonaverageshould
be able to connect to at least 95% of the other nodes.
• We call a network mostly connected () if its nodes can potentially connect to at least
half of the nodes.
• Finally, in partially connected () networks, nodes only connect to a relatively small
subset of the whole network.
Higher connectivity in the network topology leads to better resilience (availability) against
node failure, such as Denial of Service (DoS) attacks; such resilience might have, in turn,
a positive inuence on anonymity (Böhme et al. 2005). While having a fully connected
topology is better than having a very restricted network topology, such as a xed sequence
of relays, called cascade, Diaz et al. have shown a partially connected network structure—
in particular, a stratied topology—can provide better anonymity than a fully connected
network structure (Diaz et al. 2010). However, eliminating connections that might induce
security problems, such as the connection between two nodes from the same IP family
that may be easier to control by an adversary, can be benecial to anonymity. The same
holds for eliminating connections that would induce higher latency, which would, in turn,
improve the performance of the system.
(b) Connection Type. Here, we consider the direction and synchronization of connections.
As far as the direction is concerned, we consider the following options:
• A connection is unidirectional (→) if the data ow between two entities can only be in
one direction.
• A connection between two entities is bidirectional (↔) if data can ow in both directions
and the same connection is used for sending back the response to a received message.
Typically, interactive applications, such as web browsing, require bidirectional channels,
while non-interactive applications, such as email, can just close the connection as soon as
the message has been forwarded.
Bidirectional circuits have the advantage that they induce less overhead in terms of cir-
cuit construction. Unidirectional connections have the advantage that they are less vulner-
able to timing attacks, as a malicious node can only observe data owing in one direction,
which is less informative than bidirectional connections in which patterns of requests
2
In parenthesis, we dene the symbol or the keyword that is used in the comparative Tables 2, 3, 4,and5 to indicate the
corresponding characteristic.
ACM Computing Surveys, Vol. 51, No. 3, Article 51. Publication date: June 2018.

Frequently asked questions

Q1. What are the contributions in "A survey on routing in anonymous communication protocols" ?

The concept of AC systems was introduced by Chaum ( 1981 ) in 1981, with his proposal for implementing an anonymous email service that aimed at concealing who sent emails to whom. This could be explained as a response to recently revealed dragnet surveillance programs, the fact that deployed AC networks seem to become ( according to leaked documents ) a major hurdle for communications surveillance, and to somewhat increased public awareness on the threats to privacy posed by modern information and communication technologies. Of the deployed systems, the most successful example to date is the Tor network, which is used daily by about two million users ( The Tor Project 2017 ). The purpose of this survey is to provide a detailed overview of the routing characteristics of current AC systems and to discuss how their routing features impact anonymity against different types of adversaries, as well as overall performance. To this end, the authors first identify the routing characteristics that are relevant for AC protocols and provide a taxonomy for clustering the systems To select AC protocols for their examination the authors chose systems that have presentation value in terms of routing. Finally, the authors discuss the relationship between different routing decisions, and how they affect performance and scalability. Section 2 provides their taxonomy for anonymous routing and describes the various routing features and dimensions that the authors are considering for their evaluation and discusses the relationship between these routing features. Section 5 concludes the paper. In this section, the authors introduce the routing characteristics, deployability, and performance metrics considered in their taxonomy, and they discuss the relationship between these characteristics. For anonymous communication, the authors consider four building blocks that are relevant to routing in AC networks. In this section, the authors present a classification for anonymous routing protocols ( see Table 1 ). This taxonomy does not address several relevant design features of AC networks, such as probabilistic node selection for constructing circuits and security considerations for protecting routing information from different network adversaries. Nevertheless, the authors distinguish three groups of features inspired by Feeney ’ s categories: network structure, routing information, and communication model: ( 1 ) Network structure describes the characteristics of the anonymous relays, the connections between them, and the underlying network topology. In what follows, the authors describe these features in more detail, including their various sub-features and corresponding notation symbols used to denote individual feature instantiations. The authors consider first the network features that are relevant to anonymous routing. The authors note that physical topology characteristics, such as the geographical location of computers, sometimes matters in anonymous routing decisions, for example, when considering ACM Computing Surveys, Vol. 51, No. 3, Article 51. The authors consider the network as a graph in which the routers are represented by graph nodes. The authors consider a network to be fully connected ( ) when nodes can potentially connect to most ( or all ) other nodes. Here, the authors consider the direction and synchronization of connections. As far as the direction is concerned, the authors consider the following options: • A connection is unidirectional ( → ) if the data flow between two entities can only be in one direction. Further, the authors consider whether the anonymity system involves connection synchronization: • A connection is asynchronous ( ) if the establishment of connections and relaying of messages is initiated by a user without any timing coordination with other participants. The authors consider symmetry in the roles of the network entities. The authors thus first examine the overlap between the roles of end-users who initiate communications and relaying nodes. Note that the authors are not considering centralized models, because they are a single point of failure for surveillance and insecure by design. The authors now consider the information available to the entity ( or entities ) that decides on the route of a connection and how that information is made available. The authors finally consider features that describe the creation of anonymous routes. A variation of the basic sourcerouted model is found in some systems that provide receiver anonymity. In literature, other node selection strategies have been proposed, which the authors have not taken into consideration such as dynamic routing schemes using distance vector routing ( i. e., Perkins and Royer ( 1997 ) ) and link-state routing ( i. e., Moy ( 1998 ) ). An example of prioritized scheduling is when the scheduling follows an economic model, which might mitigate flooding attacks ( Grothoff 2003 ). ( c ) Node Selection. To characterize node selection, the authors consider the selection set that determines which nodes are eligible for being on the route and the selection ( probability ) distribution that describes the likelihood of each of the nodes in the selection set being chosen for a route. In other words, the authors consider parameters that are calculated in real-time to be dynamic biases, and parameters based on routing information that is unchanged until the next periodic update to be static. In addition to the routing characteristics identified before, the authors identify the following list of metrics that can be used to evaluate performance and deployability characteristics of AC protocols. If the protocols do not specify explicitly the context in which they are used, then the authors assign context/application to them mainly based on the latency and the number of intended recipients of the ACM Computing Surveys, Vol. 51, No. 3, Article 51. In this section, the authors present a categorization of AC protocols. Next, the authors discuss the AC protocols individually, starting with Mixnet protocols ( from Section 3. 1 to Section 3. 1. 3 ), and then they proceeding with Tor-related protocols ( Section 3. 2 ), Random Walk ACM Computing Surveys, Vol. 51, No. 3, Article 51. In Chaumian mixes, the mix node does not output the messages immediately upon arrival but instead collects a certain number of messages ( up to a threshold ) into a so-called batch, which introduces a delay in message transmission. Next, the authors review two variants of mix selection, one for free-route mix networks and one for mix cascades. After a mix has sent a message to the next mix, if it is not receiving a receipt within a restricted time, then it ACM Computing Surveys, Vol. 51, No. 3, Article 51. The system establishes routing paths following the free-route node selection strategy, where the mixes are selected based on their past behavior ( reputation score ). The construction of such cascade utilizes communal randomness and reputation scores provided by all of the mixes ; therefore, there is no need of a trusted central authority. The anonymity of timed mixes is vulnerable to low traffic, since if only one message arrives before the time restriction is met, the mix provides no anonymity measure for that message. For example, the two prominent remailers, namely Mixmaster ( Möller et al. 2003 ) and Mixminion ( Danezis et al. 2003 ), use timed ACM Computing Surveys, Vol. 51, No. 3, Article 51. The authors review these remailer protocols in Section 3. 1. 3. Next, the authors review some mix protocols in the literature that have been suggested for applications such as ISDN telephone, web browsing, and anonymous emails. Webmixes use an adaptation of the time-slice method introduced by ISDN mixes. Recall that the flushing ACM Computing Surveys, Vol. 51, No. 3, Article 51. However, the authors consider these two protocols with asynchronous message transmissions due to the possibility that low traffic might lead to a threshold restriction instead of a time restriction. Once the message is delivered to the entry node, the router decrypts one layer of encryption with its corresponding key and forwards it to the next hop following the instruction within the layer. Webmixes, in onion-routing, the ORs implement First-In First-Out ( FIFO ) -like forwarding strategy to provide low-latency services. The Tor network, an open-source and a free-to-use framework, consists of a large set of volunteering routers ( at the time of writing, there exist more than 7,000 routers ( The Tor Project 2017 ) ). Tor ’ s services are used daily by approximately 2,000,000 users ( The Tor Project 2017 ). To defend against this and related attacks, selecting a small set of nodes was introduced for Tor ( Wright et al. 2003 ). Next, the authors review two prominent attacks on Tor ’ s routing. The proposed extensions to the Tor routing protocol aim mostly at improving either the achieved anonymity of Tor or the performance that Tor users experience. This is important, since the Tor network consists of volunteers and is very likely to have a fraction of malicious nodes among them. Tor as a protocol where the routing decision maker has a complete view, it is worth mentioning that the unlisted relays, ACM Computing Surveys, Vol. 51, No. 3, Article 51. In this section, the authors review Random Walk protocols, where the communication is relayed randomly through the network. The authors consider a protocol a Random Walk protocol if node selection is hop-byhop routed and a random selection. To avoid repeated connections with the same set of nodes, a node has to forget about nodes it has not been connected and ACM Computing Surveys, Vol. 51, No. 3, Article 51. The further development of this concept in the past 2 decades has seen it applied to a variety of problems and scenarios, such as anonymous voting ( Sako and Kilian 1995 ; Jakobsson et al. 2002 ), Private Information Retrieval ( PIR ) ( Dingledine et al. 2000 ), censorship-resistance ( Waldman et al. 2000 ; Waldman and Mazières 2001 ), anonymous web browsing ( Goldschlag et al. 1996 ), hidden web services ( Dingledine et al. 2004 ), and many others. These results further emphasize the impact of network connectivity features for anonymous routing. The authors call a network mostly connected ( ) if its nodes can potentially connect to at least half of the nodes. Further, peer-to-peer networks are more resilient to node failures and have better availability properties. If routes are created just to send a message and no state is maintained for further exchanges, then the authors classify a protocol as message-based ( ). The authors identify several basic context/applications: namely, protocols for anonymous messaging ( ), email communication ( @ ) ; protocols for real-time communication such as telephony ( ) ; web communication, such as anonymous browsing ( ) that needs to be low-latency and microblogging ( ) that can tolerate more latency ; bulletin boards, auctions, voting, group messaging ( ) ; file-sharing ( ) ; and protocol that are used in the context of a wireless ad hoc networks ( ). Due to the temporal nature of the message-based communication, where connections are not going to be used further ( e. g., replies are not going to be sent in a short time ), setting up a circuit is unnecessary. Further, mixes shuffle ( “ mix ” ) input messages and output them in a reshuffled form. He only suggests that certain factors such as the networks topology and user ’ s trust can be used for mix node selection. Mixes that fail lead to further delays in mix networks, thus selecting reliable mix nodes can lead to better performance. Such a strategy suggests use of a non-deterministic node selection, biased toward mix nodes with high reputation scores. Web requests are sent from the users JAP proxy through the mix cascade and the cache-server and, furthermore, delivered to the destination server. Babel introduces intermix detours, where mix nodes choose a random sequence of mixes and relay the message through them before forwarding the message further to the next mix of the original route. When the message is traversing the route, at a crossover point ( the last mix in the first leg ), the SURB replaces the first leg, and the message is routed further to the intended recipient. Mixminion also suggests choosing nodes from preferably a large pool ; however, further details on the node selection strategy have not been specified in Mixminion. However, the authors distinguish whether a connection is potentially allowed between two nodes or not based on routing of the messages. The anonymous remailer Mixmaster does not discuss node management either ; however, the later implementation uses ad hoc systems, which suggests a partial view ( Danezis et al. 2003 ). For both mix cascade protocols and freeroute mix networks, the selection set varies depending on the application of the AC network and on the potential anonymity properties. Nonetheless, onion-routing is a promising design to provide a low-latency AC network, and many currently used systems can build upon this design. In the early onion-routing design, uniformly random node selection was suggested ( Syverson et al. 2001 ). As mentioned before, a further development in the routing policy is to disallow a communication to pass through two nodes within the same /16 subnet IP address. They furthermore propose some strategies to prevent the risk of such attacks, mainly by increasing communication latency ( Murdoch and Danezis 2005 ). Moreover, offering the user a tune-up option between uniformly random node selection ( for high anonymity ) and weighted random node selection with a bias toward high-bandwidth nodes ( for better performance ) has been suggested by Snader and Borisov Snader and Borisov ( 2011 ). Tor ’ s performance problems have several causes, and hence suggested improvements aim at different aspects of the Tor routing protocol. Furthermore, another solution to Tor ’ s congestion problem is to enforce avoiding congested nodes at the node selection phase ( Wang et al. 2012 ). Further inherent routing features concerning the communication model include routing type, scheduling, determinism in the node selection, and the selection set. The exceptions here are Tang and Goldberg ( 2010 ) and AlSabah et al. ( 2012 ), who suggest a prioritization at the scheduling phase in favor of interactive traffic to reduce delays that interactive users might experience. Furthermore, the node selection probability is generally weighted using static parameters, except for a few approaches that dynamically adjust weights, e. g., for balancing security versus performance ( Snader and Borisov 2011 ) and for avoiding congestion ( Wang et al. 2012 ; AlSabah et al. 2013 ). The message goes further until a node decides to send it to the final destination ( step a4 ). To prevent such type of attacks, Crowds suggests to employ static route ( a user keeps the route for a while ) such that an attacker does not have multiple routes to link to the same node ( Reiter and Rubin 1998 ). 

Q2. How does the increase in traffic affect the anonymity set size?

Expanding the bandwidth by adding a new mix cascade leads to splitting up the anonymity set size among the mix cascades and the increase in traffic would have no impact on achieving stronger anonymity. 

Q3. What are some examples of attacks against DHT-based protocols?

Examples of attacks against protocols that provide only a partial view of the network to the routing decision maker are route fingerprinting attacks (Danezis and Clayton 2006), and route bridging attacks (Danezis and Syverson 2008). 

Q4. What is the main reason why a non-deterministic node selection makes targeted surveillance harder?

Given a large number of Tor relays that are spread around the world, and since the adversary is assumed to be local, a non-deterministic node selection makes targeted surveillance harder. 

Q5. What was the first proposed system to anonymize ISDN calls?

Synchronous anonymity systems were proposed in the early 1990s by Pfitzmann et al. to anonymize ISDN telephony calls (Pfitzmann et al. 1991). 

Q6. What is the way to eliminate connections that might induce security problems?

eliminating connections that might induce security problems, such as the connection between two nodes from the same IP family that may be easier to control by an adversary, can be beneficial to anonymity. 

Q7. What are the building blocks of routing in an AC network?

These building blocks are node management, transfer/retrieval of node information to/by the routing decision maker, path selection, and forwarding or relaying; where path selection is the main design component of routing schemes for AC protocols. 

Q8. What is the main reason why a mixnet is secure against a global adversary?

Since Mixnets are secure against a global adversary the system can consist of fewer nodes than systems that are vulnerable against a local adversary. 

Q9. What is the proposed method for encrypting the address of the recipient of replies?

Chaum proposed to encrypt the address of the recipient of replies separately so that the respondent only needs to append the untraceable return address to its replies. 

Q10. What is the definition of a network-restricted subset?

It may contain a network-restricted subset ( ) of all network nodes, e.g., a subset aimed at guaranteeing the quality of the communication by, for example, avoiding congested links and nodes. 

Q11. What is the likelihood that nodes are selected for certain positions in a given route?

The likelihood that nodes are chosen for certain positions in a given route depends on the ratios of overall node bandwidths and node characteristics such as the IP addresses and whether they can be selected as entry node or as exit node. 

Q12. What is the main reason why a mixnet is vulnerable to flooding attacks?

Mixnets are also vulnerable to flooding attacks (Serjantov et al. 2003), and there needs to be a large amount of traffic entering the system to make this attack infeasible for the adversary. 

Q13. What makes Tor more resilient against DoS attacks?

Tor resilient against DoS attacks (Shirazi et al. 2015), fast, and suitable for low-latency applications, such as web browsing or instant messaging. 

Q14. What is the alternative cryptographic model for mix cascades?

An alternative cryptographic model for mix cascades is using re-encryption mixes, where mixes re-encrypt messages instead of decrypting them. 

Q15. Why are forward routes considered to have better anonymity?

Forward routes are considered to have better anonymity; one of the reasons for this is that reply blocks enable replay attacks on anonymous replies (Danezis et al. 2010). 

Q16. What is the solution to the challenge of providing real-time network data for the initiator?

One solution to this challenge would be providing real-time network data for the routing initiator, however, this solution would worsen scalability of source-routing protocols even further. 

Q17. What is the advantage of CAR over DCnets?

This allows CAR nodes and users to have a partial view of the system and the participants of the protocol, which makes CAR more scalable than DCnets.