Analysis | overview: witch attack and defense

Title: "The Sybil Attacks and one-third one-third confident: A Survey"

原文作者：Aziz Mohaisen, Joongheon Kim

原文编译：DAOrayaki

The introduction

The peer-to-peer (P2P) computing paradigm has many advantages over other traditional paradigms. For example, resources such as bandwidth, memory, and data are available to all other participating users [1]. Broadly speaking, this paradigm includes both structured and unstructured systems. Structural stacking (such as Kademlia [2] and Chord [3]) provides deterministic mechanisms for data and peer discovery, while non-structural stacking (such as Ggutella [4]) organizes peers in a random graph and uses flooding algorithms for peer and data discovery. Most popular point-to-point systems lack "central authority," which makes the paradigm resistant to failure attacks. On the other hand, the lack of such centralized authorization leads to a number of challenging security problems: most of the security services needed to protect network systems require centralized authorization of one kind or another, making these services unavailable to peer systems [5]. To make matters worse, the completely decentralized and open nature of many of these systems leads to a wide range of unknown security threats in other distributed systems, including witch attacks [6].

Witch attacks are well known in point-to-point, wired, and wireless networking environments. In its basic form, a peer representing an attacker generates as many identities as possible and acts as if she were multiple peers in the system [6], aiming to interfere with the normal behavior of the system. Id number of the attacker can generate only depends on the ability of the attacker, this ability is limited by other peers in the system response, the bandwidth required of the concurrent requests stored with witch the identity of each generation of the corresponding node routing information required memory, as well as in the case of no apparent delay handle concurrent requests the computational resources required. With the rapid growth of hardware (for example, in terms of storage capacity and processing) and the widespread spread of broadband Internet with high bandwidth rates, an attacker running even on general-purpose hardware can cause significant damage to a large system.

Witch attacks occur in many situations, as well as on point-to-point systems and other common distributed systems and paradigms that are essential to my services. Such environments include voting systems, reputation systems, routing, distributed storage, and so on. To illustrate how this attack might work in a real-world system, imagine a recommendation system built on point-to-point overlay. In such a system, the goal is to filter out information that may be of interest to the user based on the recommendations of others. In this case, an attacker who can act as multiple users by forging multiple identities (witches) can easily vote on the legitimate objects that the legitimate users vote on.

In any realistic recommendation system, it is almost guaranteed that the number of legitimate users who vote on objects will never exceed 1% of the system's total users [7]. Because of this, this kind of attack is very attractive to attackers, who are potential users in the system, trying to attack systems that offer high incentives. The online marketplace eBay, for example, uses customer recommendations to determine the reputations of its sellers and those who use its platform to sell goods, so it is tempting for such sellers to misbehave in order to gain higher reputations. The same is true in many other situations, such as peer-to-peer file sharing where content is rated by users, and bandwidth is allocated based on reputation, where even that reputation is used to determine whether the content distributed by users is good or bad. In all such cases, there is an incentive for user misbehavior, and witch attacks prove to be a powerful tool for such attackers to achieve their goals.

To defend against an attack, several forms of defense or mitigation have been attempted to defend against or limit the impact of an attack. This attack can be broadly divided into two schools of thought: centralized and decentralized (i.e., distributed) defense. In centralized defense [6, 8, 9, 10], a central authority is responsible for verifying the identity of each user in the system. While this defense is somewhat effective against attacks, it makes certain assumptions about the system, some of which are not easy to implement in a decentralized point-to-point system. First, as the name and operation description suggest, such systems require centralized permissions, which many of them may not be able to afford for security and functionality reasons. Moreover, even if such a centralized authority exists, some credentials associated with the user in the system are required in order to match those credentials with the user's digital identity: in many cases, obtaining such credentials can be a challenging task.

Decentralized defense, on the other hand, includes, but is not limited to, work in [11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 7, 21, 22, 23] that does not require centralized authority in the system and is carefully designed for decentralized point-to-point systems. At the core of their operations, such defenses weigh the collaboration between users in the system to admit or reject users who may be attackers. A user's acceptance or rejection is based on the credentials associated with the user, as in the case of encrypted distributed defense, or the network attributes of a legitimate honest user, as in the case of witch defense using a social graph. In either solution, the ultimate goal of defense is to simulate the power of central authority in a decentralized manner and use that power to detect witches and honest nodes.

Another classification of defense might be based on how it operates. In previous work, therefore, the witch defense defense, which can be divided into use trusted certification is usually generated for honest users, and according to the public key to verify the trusted institutions, produce cost - the user some cost penalty, to limit the number of their available resources and reduce their wrongdoing, and the witch defense based on social network.

Such defences vary greatly in their assumptions, the type of network used, the guarantees provided, and the costs incurred. To this end, this paper mainly reviews, summarizes, compares and shows the shortcomings of the existing methods of witch defense.

2. Witches attack models and targets

In this section, the attacker model commonly assumed in most witch attack studies and defenses is explained in detail, and the targets of attackers and defenses are further explored.

2.1 & have spent Problem statements and models

The problem is expressed as the fact that a single user in the system has an ability in the system that allows her to act as if she had multiple avatars. This is problematic for so many applications, because the correctness of such applications depends on the behavior of peers, the number of them, and the willingness to collaborate honestly in the system. However, the witch identity of a single attacker who tries to bias the overall behavior of the system does not satisfy such a system goal.

The official hallmark of an attacker is the number of false identities she can inject into the system. The attacker's motivation is to maximize the number of false identities. The value and significance of the number of identities that an attacker generates and injects into the system depends on the application and varies from application to application. For example, to attack a recommendation system, it would be sufficient to make the sum of matches of 1% of honest users in the system a false identity. Since it is generally observed that even for a very popular object in the system, only 1% of the honest nodes vote for it, this is especially one that is enough to skew the behavior of the system and outperform the honest nodes in the system. That is, by having more single identities than the exact number of honest nodes that an object in the system votes for, the witch attacker will be able to outvote the votes of the honest nodes.

On the other hand, in other systems, such as hybrid networks used for communications anonymization (such as the Tor network), even a small number of witch identities can seriously undermine the promise in the system. Theoretically, the compromise between two nodes on the circuit is sufficient to identify the sender and receiver of communication on such a hybrid network [24]. On the other hand, the compromise of enough identities in a network would allow an attacker to monitor any number of circuits. Other applications where the number of identities themselves is important include attacks on file-sharing systems [25], and there are many examples.

Figure 1. Several different ways witches defend themselves in a P2P overlay environment

C/CA stands for Centralized Certification Body

D/C stands for Decentralized Encryption primitives

T/D stands for Trusted Device Authentication

IP/T stands for IP detection

R/C stands for Cost Cycle Based Detection

S/G stands for Social Graph Based Defense

2.2 & have spent The goals and success criteria of witch defense

The ultimate goal of witch defense is to eliminate witch attacks by detecting and isolating witch identities in the overlay network, or nodes that are equally capable of spawning these witch identities. However, the ultimate goal is not always able to achieve, because most of the defense, besides based on centralized credible certification scheme, most of the defense in the detection mechanism is the possibility of errors and omission, might tolerate certain witch node, falsely report other honest nodes at the same time, as we see in the correct false positives and false negatives. False positive reports refer to witch nodes being reported as honest nodes. On the other hand, false negative honest nodes are reported as witch nodes. The realistic and realistically achievable goal of defense mechanisms is to minimize false negatives.  

2.3 & have spent The depth of the witch's defense

In addition to the broad categories of centralized and decentralized defenses shown in Section 1, the witch defenses investigated in this article include two broad categories of techniques: trusted authentication and resource testing categories, as shown in Figure 1. In the Trusted Authentication list, we comb through the work involved in using a centralized authentication authority, decentralized cryptographic primitives, or trusted devices.

3. Use trusted authentication

The trusted authentication method can be said to be one of the most popular methods at the moment, because Douceur [6] has proved its potential to eliminate Sybil attacks [26]. In the traditional form of this approach, centralized authorization (CA) is used to ensure that the identity assigned to each peer is unique and legitimate by matching these identities with pre-assigned credentials. These credentials may include encryption keys, synchronous random strings typically generated by a one-time password generator (OTP), or digital certificates issued by centralized authentication.

Although the above traditional form of a centralized certificate authority is well defined in the literature, the distributed certificate scheme has been defined by applying cryptographic primitives that apply to a distributed multi-party model that enables collaboration between supposedly honest peers to justify other peers joining the overlaid.

3.1 & have spent Centralized certification bodies

Centralized trusted authentication may be the only way to eliminate witch attacks [6]. In the context of P2P overlay, there has been some work on using a centralized certificate authority for certificate generation, allocation, and validation. For example, a study using the social graph explained in Section 5 and using public key cryptography as a building block assumes the authenticity of the user's public key through certificates assigned to the user through centralized authorization in the offline phase [19]. Other examples of schemes utilizing CCA-based approaches include the work in [8], [9], [6] and [10].

3.2 & have spent Cryptographic primitives

Recently, there has been a flurry of work on cryptographic primitives [11, 12, 13, 14, 15]. These primitives are intended to provide the infrastructure for validating peers in order to make witch attacks harder to occur by having only legitimate peers participate in overwriting. Typically, these efforts attempt to leverage the public key infrastructure (PKI) in a decentralized manner and use threshold encryption components (for example, secret sharing and threshold signature) to ensure collaboration between so-called honest users to verify the operation time of joining peers overwritten on them. Interestingly, the explicitly stated motivation goes beyond some of these primitives (for example, [11, 12, 13, 14]), because many non-cryptographic protocols in the literature assume the existence of authentication systems to cover legitimate users (for example, SybilGuard and SybilLimit). Therefore, the encryption method is designed to ensure the successful operation of such a protocol.

3.3 & have spent The trusted device

Similar to the idea of trusted authentication, some research and practice suggest the use of trusted devices or trusted modules to store certificates, keys, or authentication strings previously assigned to users by a centralized agency. This device is largely difficult to obtain due to its potential high price and can therefore be used to limit opportunities for witch attackers. Examples of this mechanism are presented in [27] and [28], although later work has been on wireless sensor networks. In theory, these defenses might work when an attacker is intent on acquiring as many witch identities as possible. However, in the case of anonymity (such as Tor) and recommendation systems, these defences are outdated, given that a small number of witches can do a lot of damage.

4. Resource testing

In addition to the resource testing method used to defend against witch attacks, the basic idea is to check whether a set of identities associated with supposedly different users has enough resources to match the number of identities. These resources may include computing power, bandwidth, memory, IP addresses, and even trust credentials. Although Douceur has demonstrated that the idea of resource testing is ineffective [6], some researchers consider it to be a minimal defense. That is, it doesn't eliminate attacks completely, but it does make witch attacks harder to do.

In theory, most schemes in such defenses limit the number of witch identities to fewer than they would be without defenses. In practice, however, even a smaller number of witch identities is enough to threaten the usability and security of many systems. For example, as mentioned earlier, anonymity in an anonymous system (such as TOR) depends on two nodes per circuit. In addition, 1% of fake identities in online reputation systems are enough to vote for legitimate nodes.

4.1 IP  test

Common testing scenarios involve testing the IP addresses of peers, trying to determine their location and match it with their activities. In particular, if there is a large amount of activity from the same particular geographical area, it is likely that some of that activity is due to witch status. Furthermore, the assumption in such works is to get IP addresses for different geographic areas, but that's not cheap. For example, Friedman et al. [29] introduced TARZAN, where a node's IP address is tested based on its geographic location within a particular autonomous system. Cornelli et al. presented similar results in [30] and [30].

The main assumption of this work is that IP addresses are difficult to obtain over a wide geographical area. However, with recent indications of the existence of large botnets [31], where infected hosts are controlled by a single administrative entity and reside in different autonomous systems, it is certain that this defense mechanism is useless.

4.2 & have spent The cost of circulation

Some studies and practices suggest recurrent costs as a means of defending against witch attacks. In particular, computational puzzles [16, 32] and Turing tests (e.g. CAPTCHA [33]) have been suggested as solutions. However, by the same token, IP testing won't work against attackers controlling botnets, and neither will these cost-based schemes. In addition, for solutions like CAPTCHA, it has been shown that a Sybil attacker might publish CAPTCHA tests on controlled sites for users to test to access services provided by the attacker.

In addition, some versions of CAPTCHA are vulnerable to certain image processing attacks [34].

5. Social Graph Based Defense

While most of the previous solutions to the problem of witch attacks in distributed systems have limitations and disadvantages, the social network-based witch defense attempts to overcome these problems in an elegant way. First, social network-based witch defenses are mostly decentralized solutions to the problem of witch attacks, which means these designs operate without any central authorization -- an important feature in most distributed systems. This decentralized model of operation is made easier by the random walk theory, which is the most common component of these defenses. Second, these defenses leverage trust in social links between social nodes, making collaboration between honest nodes possible and easy. Third and finally, these defenses have been demonstrated in several studies to be practical and effective against witch attacks at low cost, and they are now being further developed as components for many services, including distributed hash tables (DHT), witch-proof voting, and for mobile network routing.

Although they differ greatly in design details and operation, all social network-based witch defenses share two common assumptions: algorithmic attributes that are fast mixing attributes and trust. First, these defenses are based on the "quick-mixing" nature of the social graph. Unofficially, the fast-mixing nature of a social graph means that the "honest" nodes in such a graph mesh well, and the honest region does not contain a sparse cut -- a cut that connects two large subsets of honest nodes to some social network. For simplicity, the fast mixing nature of the social graph means that a random walk from any node in the social graph will be very close to the smooth distribution of the Markov chain (MC) defined on the graph a few steps later. In a million-node network, the recommended number of such steps is 10 to 15.

The second common assumption in this defensive vein is trust. In particular, all of these defenses assume good trust values in the underlying social graph, for example, as indicated by face-to-face interactions between nodes. This particular assumption is necessary in order to infer how easy it is for any number of attackers' social links to penetrate a social network. Although the operation of the witch defense for correctly identifying the "honest" nodes in the social graph is guaranteed by the fast mixing assumption, and the construction of corresponding schemes using this algorithmic attribute, the ability to identify the witch node is guaranteed only if the attacker, or the attacker collectively, assumes It controls some of the links between itself and other honest nodes in the social graph (such links are called attack edges).

Here we comb through some of the most widely cited and noted methods of social network-based witch defense.

5.1 & have spent Sybil Guard

The design of Sybilguard, thanks to Yu et al. [19, 20], uses the fast mixing properties of trust ownership of social networks to detect witch nodes. Technically, SybilGuard consists of two phases: the initialization phase and the online detection phase. During the initialization phase, each node builds its routing table as a random permutation of its neighbors for input and output edge pairs. Next, each node launches a random walk of length w = O (square root n log n) and propagates it to its neighbors according to a routing table built using random permutation. Each node on the random walk path registers as a witness to the initiator of the random walk, and then acts as a witness to the node when it becomes a suspect. In addition, taking advantage of the retrospective nature of the random walk, each initiator of the random walk receives a list of "witnesses" (that is, nodes that register the initiator's public key and are located on the path built by the initiator's random walk).

In the online phase, the verifier determines whether the suspect is honest, as shown below. First, the suspect sends the address of the "witness" on his random route to the verifier. Therefore, the verifier compares the witness list with its verifier route list. If there is no intersection between the two sets (a very likely event), the verifier will abort and reject the suspect. Otherwise, the verifier will go ahead and ask to compare nodes on the intersection between two sets to verify that the suspect is using their registered public key. If the suspect is validated by a cross-node, the verifier accepts the suspect or marks him as a witch node.

5.2 & have spent Sybil Limit

Unlike SybilGuard, which uses a single long random walk, SybilLimit recommends using multiple shorter random walks. In addition, unlike SybilGuard (where the verifier and suspect's public keys are registered at nodes in the social graph), SybilLimit [18] recommends that such keys be registered at the edges of the social graph. Sybillimit consists of an initialization phase and an online validation phase. During the initialization phase, each node constructs its routing table using the same method described in SybilGuard and performs a random walk of r = O (square root of n), each of length w = O (log n) where O (log n) is 10 to 15 social graphs of mixed time social graphs, The theoretical assumption is sufficient to sample nodes from distributions that are very close to statistical distributions. Unlike SybilGuard, all nodes on the random route path are used to register the public key of the initiator of the random walk, and the last edge of each walk in r random walks is used to register the public key of the initiator node (each such edge is called the tail). In particular, the public key of the walk initiator is registered on the last node under the last edge reached by the random walk. Also taking advantage of the traceability of random routing, witnesses who register the public key of the initiator node (which can be a suspect or verifier) return their identity to the node. Each node in the social graph performs the same process, and each node that initiates the registration random walk collects a set of witnesses (or validation nodes).

In the online phase, SybilLimit is also similar to SybilGuard, where the suspect sends the witness's identifier and address to the verifier node, which compares witnesses in the suspect list in an attempt to find conflicts. If there is a conflict between two sets on the verifier side, the verifier will ask witnesses with common identities in both sets to verify the identity of the suspect, and decide whether to accept or reject the suspect based on the results of this process. If there is no intersection between the two (which is highly unlikely to happen) the verifier terminates and rejects the suspect by labeling him as an attacker.

The provable guarantee for inferring SybilLimit has the same main components as that for Sybil Guard. In particular, it is assumed that the random walk length W is the mixing time of the social graph, then the last node chosen in such a random walk is based on a stationary distribution.

Furthermore, the last edge in the random walk is selected "almost" uniformly and randomly from among the edges in the social graph. In addition, assuming R = O (square root of N), if the hidden constant R0 is correctly selected (where R = R0 × square root of N), then the intersection between the sample edges of the verifier and suspect will exist with overwhelming probability. The authors call this condition the "intersection" condition, which is used to guarantee the high probability intersection of the nodes' random walk in the honest region. As with SybilGuard, assuming G attacker edges, an attacker is allowed to register the public key of his witch identity on a maximum of GWR = O (G × square root N × log N) tails (called the Contamination Tail). In this case, each additional edge introduces an additional O (log n) witch identity (assuming the attacker uses the best attack strategy by registering a different public key with a different Sybil identity at each potentially contaminated tail).

The security of SybilLimit also relies heavily on W. Since there is no mechanism to estimate the exact value of the parameters, either underestimating or overestimating them can be problematic, as shown above. Sybillimit also provides a "benchmark technique" for estimating this parameter, which also does not provide any provable assurance about the quality of the parameter estimation. Finally, as long as G = O (n/log n), the Sybil Limit guarantees the number of witch identities introduced per attack edge. Note that neither SybilGuard nor SybilLimit requires any global knowledge of the social network they operate and can be implemented in a completely decentralized manner.

5.3 & have spent Sybil Infer

SybilInfer uses a probabilistic model defined on a random walk (called a trajectory) to deduce the truth about the set of nodes X that generate such a trajectory. The basic assumptions in SybilInfer are that each node has a global view and knowledge of the social network, that the network is quickly blended, and that the node that initiates SybilInfer is an honest node. Technically, SybilInfer tries to eventually label the different nodes in the graph as honest nodes or witch nodes. In SybilInfer, each node in a network of n nodes performs s walks, so the total number of walks in the common track is s × n. Each trace in these trails is made up of the first node (the initiator of the random walk) and the last node in the random walk (the tail). Unlike the uniform (beyond node degree) transfer probability used in SybilGuard and SybilLimit, SybilInfer defines a uniform transfer matrix at the nodes to penalize nodes with higher heights. SybilInfer operation's ultimate goal is to calculate the probability P (X = could | T); In other words, calculate the probability that a set of nodes X for a given trajectory T is honest. This probability is calculated using Bayes' theorem.

SybilInfer uses the technology to sample the honesty configuration, which was originally used to determine the honesty of a set of nodes from a trace. This sampling is done using the Metropolis-Hasting algorithm, which first considers a collection X0 and modifies the collection once by removing or adding nodes to the collection: At each time, and with a probability of filling a new node from X¯0, X is added to X0 such that X '= X0 [X or nodes in X0 are premoved with a probability to be removed. The process executes n× log n rounds to get a good sample independent of X0.

5.4 & have spent Sum Up

Unlike Sybilguard and Sybillimit, which are generic node access issues and are centralized in the sense that a single node does not need to carry global information about the social graph, and Sybil-Infer, which is used to Infer node honesty, SumUp [35] attempted to address the witch-attack problem in the context of voting convergence. In this case, a node called the vote collector wants to collect votes from other nodes in the network in an anti-witch fashion. That is, for a given number of object votes, the vote collector wants to increase the percentage of votes received by honest nodes, reduce the number of votes cast by attackers through their attack edges, and identify attackers when they repeatedly misbehave. At the heart of SUMUP, the link capacity allocation mechanism is used to adaptively allocate capacity to links in the trust owning social graph and to limit the number of votes propagated from the voter's side to the poll collector. SumUp's adaptive vote-flow mechanism uses two observations of a traditional online voting system: a small number of users in the system vote on a single object, and -- if such a voting system is implemented on top of a social graph -- congestion occurs only on the chain close to the poll collector. Therefore, SumUp suggests distributing a large number of tickets on different links in the social graph based on their distance from the voting collector (based on some level calculated using the breadth-first search algorithm).

One obvious appeal of this technique is its high computational requirements: the running time of a typical algorithm, such as Ford-Fulkerson, would require an order of operating the number of edges to collect the votes of a single voter. Its authors further propose a heuristic approach that uses only an order of the number of steps in the graph diameter, where each node greedily selects a higher-level node through which votes are connected using non-zero capacity and propagated until a vote collector is reached. At any given time, given that the greedy step may not result in non-zero capacity, each node is allowed to explore other nodes to find the same or lower level path.

5.5 the GateKeeper

Gatekeeper [21] borrows the tools of SumUp and SybilLimit for efficient operation. In particular, it attempts to improve the performance of SybilLimit by incorporating SumUp's ticket distribution component. Unlike in SUMUP, where nodes are accepted via a non-zero path from voter to collector, Gatekeeper, as mentioned earlier, only considers the "ticket allocation" phase of SUMUP, where tickets are used by the access controller to accept nodes. Such tickets are propagated from the controller to all nodes in the same manner as in SUMUP. However, in order to limit the opportunity for attackers to obtain more tickets and reduce their overall advantage, the controller in Gatekeeper randomly selects M different random nodes. These nodes are referred to as "vantage nodes" if and only if the suspect node receives Fadmitm tickets from different vantage points (where fadmit is the score of randomly selected vantage points; It is only allowed if 0.2 is used in Gatekeeper. Therefore, once a node is allowed by a small part of this vantage point, it will be allowed. To prevent a double-flower attack, Gate Keeper recommends using the ticket's path to an encrypted signature chain (propagated to the controller).

5.6 & have spent Other Social Network-based & NBSP; DHT

Sprout [36] is another DHT routing protocol that uses social links with trusted social graphs to route information to users operating on a social network. Sprout is actually built on top of Chord [3], where additional links (routing table entries) are added to other users in the social network of any given node who are online at all times. By doing so, Sprout claims to improve the reliability and load distribution of Chord itself.

Whanau originally proposed it in [23], where the work in [22] includes further analysis and demonstration of performance and security and the implementation and demonstration of end-to-end assurance.

In [1], the author uses a bootchart -- showing the tree of relationships introduced into DHT to defend against witch attacks. By modifying the operation of the interested DHT Chord so that each node returns the address of all nodes it knows (including the entry point), the authors devised several strategies to reduce the impact of witch attacks. Unlike the original Chord, which used the proximity on Chord as a routing (query) metric, this solution takes into account a variety of routing strategies, including multiplicity, blending, and Zig-Zag. The authors have experimented to show that when this policy is run under witch attacks, it can be used to perform anti-witch DHT lookups more efficiently, and that it will require fewer queries than a normal Chord design would require.

Mobid's [37] design is a social network-based witch defense that claims to provide strong defenses for mobile environments, whereas existing defenses are primarily designed for peer-to-peer networks and are based on random walk theory. In addition, Mobid uses the number of mediators (a graph-theoretic metric in the social graph) to determine the goodness of nodes to defend against witch attacks. However, this work does not appear to offer any verifiable assurances. Comparison of the various programmes with others in the literature is shown in Tables 1 and 2.

Conclusion 6.

P2P overlay networks are highly vulnerable to witch attacks, and due to the nature of P2P overlay, solving the problem of witch attacks is more difficult than ever: it discourages the centralized certification authority needed for secure enforcement, and sometimes even centralized certification authorities are not present in P2P overlay designs. In this article, the different methods used to defend against witch attacks in P2P overlay are reviewed. The assumptions, characteristics and shortcomings of different defenses are compared.

References:

[1] George Danezis, Chris Lesniewski-laas, M. Frans Kaashoek, and Ross Anderson. Sybil-resistant dht

Routing. In Esorics, Lecture Notes In Computer Science, Pages 305 -- 318, Berlin, Heidelberg, 2005. Springer.

[2] Petar Maymounkov and David Mazi`eres. Kademlia: A peer-to-peer information system based on the xor metric. In Peter Druschel, M. Frans Kaashoek, And Antony I. T. Rowstron, editors, IPTPS, Lecture Notes in Computer Science, pages 53 -- 65, Berlin, Heidelberg, 2002. Springer.

[3] Ion Stoica, Robert Morris, David R. Karger, M. Frans Kaashoek, and Hari Balakrishnan. Chord: A Scalable Peer-to-Peer Lookup Service for Internet Applications. In Sigcomm, Pages 149 -- 160, New York, NY, USA 2001. ACM.

[4] Yong Wang, Xiao chun Yun, and Yifei Li. Analyzing the characteristics of gnutella overlays. In ITNG, Pages 1095 -- 1100, Washington, DC, USA, 2007. IEEE Computer Society.

[5] Vivek Pathak and Liviu Iftode. Byzantine fault tolerant public key authentication in peer-to-peer

Systems. Computer Networks, 50(4):579 -- 596, 2006.

[6] John Douceur and Judith S. Donath. The Sybil Attack. In IPDPS, Pages 251 -- 260, Washington, DC, USA, 2002. IEEE.

[7] Haifeng Yu, Chenwei Shi, Michael Kaminsky, Phillip B. Gibbons, and Feng Xiao. Dsybil: Optimal Sybil - Resistance for Recommendation Systems. In IEEE Symposium on Security and Privacy, Pages 283 -- 298 Washington, DC, USA, May 2009. IEEE Computer Society.

[8] Miguel Castro, Peter Druschel, Ayalvadi J. Ganesh, Antony I. T. Rowstron, and Dan S. Wallach. Secure routing for structured peer-to-peer overlay networks. In OSDI, Berkeley, CA, USA, 2002. USENIX Association.

[9] Atul Adya, William J. Bolosky, Miguel Castro, Gerald Cermak, Ronnie Chaiken, John R. Douceur, Jon Howell, Jacob R. Lorch, Marvin Theimer, and Roger Wattenhofer. Farsite: Federated, available, and reliable storage for an incompletely trusted environment. In OSDI, New York, NY, USA, 2002. USENIX Association.

[10] Jonathan Ledlie and Margo I. Seltzer. Distributed, secure load balancing with skew, Everything considered and churn. In Infocom, Pages 1419 -- 1430, Washington, DC, USA, 2005. IEEE.

[11] Francois Lesueur, Fran, Francois Lesueur, Ludovic M´e, And Val´ Erie Viet Triem Tong. An Efficient Distributed PKI for Structured P2P Networks. In Proceeding of P2P. Pages 1-10, Washington, DC, USA, 2009. IEEE

Computer Society.

[12] FRAN COIS LESUEUR, LUDOVIC M´ E, And Val´ Erie Viet Triem Tong. A Distributed Certification System for Structured P2P Networks. In David Hausheer and Editors, Aims, Volume 5127 of Lecture Notes in Computer Science, Pages 40 -- 52, Berlin, Heidelberg, 2008. Springer.

FRAN COIS LESUEUR, LUDOVIC M´ E, FRAN COIS LESUEUR, LUDOVIC M´ E, And Val´ Erie Viet Triem Tong. A SYBIL-Resistant Admission Control Coupling Sybilguard with Distributed Certification. In Wetice, Pages 105 -- 110, Washington, DC, USA, 2008. IEEE Computer Society.

[14] Fran Francois Lesueur, Ludovic M´e, and Val´ Erie Viet Triem Tong. A Sybilproof Distributed Identity

Management for P2P Networks. In ISCC, Pages 246 -- 253, Washington, DC, USA, 2008. IEEE.

[15] Agapios Avramidis, Panayiotis Kotzanikolaou, and Christos Douligeris. Chord-pki: Embedding a

public key infrastructure into the chord overlay network. In Javier Lopez, Pierangela Samarati,

And Josep L. Ferrer, Editors, EuroPKI, Volume 4582 of Lecture Notes in Computer Science, Pages 354 -- 361, Berlin, Heidelberg, 2007. Springer.

[16] Nikita Borisov. Computational puzzles as sybil defenses. In Alberto Montresor, Adam Wierzbicki,

And Nahid Shahmehri, editors, peer-to-peer Computing, pages 171 -- 176, Washington, DC,

USA, 2006. IEEE Computer Society.

[17] Haifeng Yu, Phillip B. Gibbons, and Michael Kaminsky. Toward an optimal social network

defense against sybil attacks. In Indranil Gupta and Roger Wattenhofer, editors, PODC, pages

376-377. ACM, 2007.

[18] Haifeng Yu, Phillip B. Gibbons, Michael Kaminsky, and Feng Xiao. Sybillimit: A Near-Optimal Social Network Defense Against Sybil Attacks. In IEEE Symposium on Security and Privacy, Pages 3 -- 17 Washington, DC, USA, 2008. IEEE Computer Society.

[19] Haifeng Yu, Michael Kaminsky, Phillip B. Gibbons, and Abraham Flaxman. SybilGuard: defending against sybil attacks via social networks. In Luigi Rizzo, Thomas E. Anderson, and Nick McKeown, editors, SIGCOMM, Pages 267 -- 278, New York, NY, USA, 2006. ACM.

[20] Haifeng Yu, Michael Kaminsky, Phillip B. Gibbons, and Abraham D. Flaxman. Sybilguard: Defending Sybil attacks via social networks. IEEE/ACM Trans. NetW., 16(3): 576-589, 2008.

[21] Nguyen Tran, Jinyang Li, Lakshminarayanan Subramanian, and Sherman S.M. Chow. Optimal sybil-resilient node admission control. In The 30th IEEE International Conference on Computer Communications (INFOCOM 2011), Shanghai, P.R. China, 2011. IEEE.

Chris Lesniewski-Lass and M. Frans Kaashoek. Wh¯anau: A Sybil -proof Distributed Hash Table. In

7th USENIX Symposium on Network Design, Pages 3 -- 17, Berkeley, CA, USA,

2010. USENIX Association.

[23] C. Lesniewski-Laas. A Sybil-proof one-hop DHT. In Proceedings of the 1st workshop on Social network systems, Pages 19 -- 24, New York, NY, USA, 2008. ACM.

[24] Paul F. Syverson, David M. Goldschlag, and Michael G. Reed. Anonymous connections and

Onion Routing. In IEEE Symposium on Security and Privacy, Pages 44 -- 54, Washington, DC, USA,

1997. IEEE Computer Society.

[25] Peng Wang, James Tyra, Eric Chan-tin, Tyson Malchow, Denis Foo Kune, and Yongdae Kim.

Attacking the kad network, 2009.

[26] B.N. Levine, C. Shields, and N.B. Margolin. A survey of solutions to the sybil attack. Technical

report, University of Massachusetts Amherst, Amherst, MA, 2006.

[27] Rodrigo Rodrigues, Barbara Liskov, The Design of a Robust Peer-to-Peer System. In 10th ACM Sigops European Workshop, Pages 1 -- 10, New York, NY, USA, 2002. ACM.

[28] James Newsome, Elaine Shi, Dawn Song, and Adrian Perrig. The sybil attack in sensor networks: analysis & One-third confident. In IPSN "04: Proceedings of the 3rd International Symposium on Information Processing in Sensor Networks, Pages 259 -- 268, New York, NY, USA, 2004. ACM.

[29] Michael J. Freedman and Robert Morris. Tarzan: a peer-to-peer anonymizing network layer. In

Vijayalakshmi Atluri, Editor, ACM Conference on Computer and Communications Security, Pages 193 -- 206, Washington, DC, USA, 2002. ACM.

[30] Fabrizio Cornelli, Ernesto Damiani, Sabrina De Capitani di Vimercati, Stefano Paraboschi, Pierangela Samarati. Choosing a Reputable Servents in a P2P Network. In WWW, Pages 376 -- 386, New York, NY, USA 2002. ACM.

[31] Brent ByungHoon Kang, Eric Chan-Tin, Christopher P. Lee, James Tyra, Hun Jeong Kang, Chris Nunnery, Zachariah Wadler, Greg Sinclair, Nicholas Hopper, David Dagon, and Yongdae Kim. Towards complete node enumeration in a peer-to-peer botnet. In Wanqing Li, Willy Susilo, Udaya Kiran Tupakula, Reihaneh Safavi-Naini, and Vijay Varadharajan, editors, Asiaccs, pages 23 -- 34, New York, NY, USA, 2009. ACM.

[32] Frank Li, Prateek Mittal, Matthew Caesar, and Nikita Borisov. Sybilcontrol: practical sybil

defense with computational puzzles. In Proceedings of the seventh ACM workshop on Scalable trusted computing, Pages 67-78. ACM, 2012.

[33] Luis von Ahn, Manuel Blum, Nicholas J. Hopper and John Langford. Captcha: Using hard ai problems for security. In Eli Biham, editor, EUROCRYPT, volume 2656 of Lecture Notes in Computer Science, Pages 294 -- 311, Berlin, Heidelberg, 2003. Springer.

[34] Jeff Yan and Ahmad Salah El Ahmad. Breaking visual captchas with naive pattern recognition algorithms. In ACSAC, Pages 279 -- 291, Washington, DC, USA, 2007. IEEE Computer Society.

[35] N. Tran, B. Min, J. Li, and L. Subramanian. Sybil-resilient online content voting. In NSDI, Berkeley, CA, USA, 2009. USENIX.

[36] Sergio Marti, Prasanna Ganesan, and Hector Garcia-Molina. DHT Routing Using Social Links. In IPTPS, Pages 100 -- 111, Washington, DC, USA, 2004. IEEE.

[37] Daniele Quercia and Stephen Hailes. Sybil attacks against mobile users: friends and foes to the

Rescue.in Infocom "10: Proceedings of the 29th Conference on Information Communications, Pages

336-340, Piscataway, NJ, USA, 2010. IEEE Press.

[38] Abedelaziz Mohaisen, Aaram Yun, and Yongdae Kim. Measuring the mixing time of social graphs. In Proceedings of the 10th annual conference on Internet Measurement, IMC "10, Pages 383 -- 389

New York, NY, USA, 2010. ACM.

[39] Bimal Viswanath, Ansley Post, Krishna P. Gummadi, and Alan Mislove. An analysis of social network-based sybil defenses. In SIGCOMM, New York, NY, USA, August 2010. ACM.

[40] George Danezis and Prateek Mittal. SybilInfer: Detecting sybil nodes using social networks. In

The 16th Annual Network & Distributed System Security Conference, Lecture Notes in Computer Science, Berlin, Heidelberg, 2009. Springer-Verlag.

[41] Abedelaziz Mohaisen, Huy Tran, Nicholas Hopper, and Yongdae Kim. Understanding social networks properties for trustworthy computing. In ICDCS Workshops, Pages 154-159. IEEE, 2011.

[42] Abedelaziz Mohaisen and Scott Hollenbeck. Improving social network-based sybil defenses by augmenting social graphs. In WISA, 2013.

[43] Abedelaziz Mohaisen, Nicholas Hopper, and Yongdae Kim. Keep your friends close: Infocom, Pages 1943 -- 1951. IEEE, 2011. Incorporated trust into social network-based sybil confident.

[23] Haifeng Yu. Sybil safe knowledge via social networks: A tutorial and survey. ACM SIGACT News, 42(3):80 -- 101, 2011.

The original link

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia