PROOF-OF-REPUTATION: AN ALTERNATIVE CONSENSUS MECHANISM FOR BLOCKCHAIN SYSTEMS

International Journal of Network Security & Its Applications (IJNSA) Vol.13, No.4, July 2021
DOI: 10.5121/ijnsa.2021.13403 23
PROOF-OF-REPUTATION: AN
ALTERNATIVE CONSENSUS MECHANISM
FOR BLOCKCHAIN SYSTEMS
Oladotun Aluko1
and Anton Kolonin2
1
Novosibirsk State University, Novosibirsk, Russia
2
Aigents Group, Novosibirsk, Russia
ABSTRACT
Blockchains combine other technologies, such as cryptography, networking, and incentive mechanisms, to
enable the creation, validation, and recording of transactions between participating nodes. A consensus
algorithm is used in a blockchain system to determine the shared state among distributed nodes. An
important component underlying any blockchain-based system is its consensus mechanism, which
principally determines the performance and security of the overall system. As the nature of peer-to-
peer(P2P) networks is open and dynamic, the security risk within that environment is greatly increased
mostly because nodes can join and leave the network at will. Thus, it is important to have a system that can
check against malicious behaviour. In this work, we propose a reputation-based consensus mechanism for
blockchain-based systems, Proof-of-Reputation(PoR) where the nodes with the highest reputation values
eventually become part of a consensus group that determines the state of the blockchain.
KEYWORDS
Consensus Mechanism, Distributed Ledger Technology, Blockchain, Reputation System, Social Computing.
1. INTRODUCTION
Distributed consensus among nodes that are geographically distributed has been a widely studied
research topic in distributed systems. However, with the introduction of blockchain, it has gotten
even more attention because blockchains are a type of distributed system. Most blockchain-based
systems aimed at various application domains with a variety of unique requirements have
introduced a corresponding consensus mechanism tailored to their specific uses. As a result,
several consensus algorithms with varying properties and capabilities have emerged.
A blockchain system, at its core, is a distributed system that uses a consensus algorithm to
determine the shared state among distributed nodes. This shared state, known as a chain, is a
public or private record of all transactions or digital events that have been created and shared
among participating nodes [1]. The primary goal of any blockchain-based system is to maintain a
live decentralized transaction ledger while defending against malicious Byzantine actors who
may attempt to game the system. The reliability and integrity of the entire blockchain system are
heavily reliant on the consensus model used. The applicability of any consensus mechanism is
based on three key properties: safety, liveness, and fault tolerance [2].
Aside from appropriate integrity criteria with which transactions are verified, the successful
operation of a blockchain system is also dependent on many other key aspects such as the
correctness of technical protocols, strong cryptographic mechanisms for managing participant

24
identity and checking transaction integrity, and an incentive to motivate nodes who participate in
the community while maintaining an honest behavior over time. Proof-based consensus
mechanisms are the most common approach to consensus in blockchain systems. The basic idea
behind a proof-based consensus algorithm is that, among the many nodes that join the network,
the node that performs sufficient testing is given the right to add a new block to the chain and is
rewarded. [3]. A large number of these methods are still vulnerable to the games of the
participants on the network. PoW (Proof of Work) consensus algorithm, which powers the
Bitcoin cryptocurrency, is the most popular where each participant in the system votes by the
total amount of computing power that the participant controls at the time of the vote. The obvious
disadvantage of this approach is that anyone with the most computing power can essentially take
over a significant portion of the system. In addition, it has been known to consume a significant
amount of resources [4].
As peer-to-peer (P2P) networks are open and dynamic in nature, the security risk within that
environment is greatly increased, owing to the fact that nodes can join and leave the network at
will. As a result, it is critical to have a system in place that can detect malicious behavior. Using
community-based reputations is one way to reduce the risks associated with this type of open
community. Historically, reputation systems have been employed to facilitate trust between
entities [5].
The goal of this work is to create an alternative consensus mechanism for blockchain networks.
This consensus mechanism is based on the reputation of network participants. It uses the
interaction of nodes in the network over time to determine the amount of reputation associated
with each node in the network. It then uses the reputation to determine a set of consensus nodes
responsible for maintaining the network's shared state, as well as the calculation of new
reputation values as interactions between nodes in the network progress over time.
The main contributions of this work are described as follows:
● First, in our reputation-based consensus mechanism, a node's reputation is calculated by
blending together a normalized set of ratings and the corresponding reputation values of
the node providing the rating at a given point in time, rather than simply the value of the
direct rating given by other nodes. A node's behavior influences its overall reputation
value.
● Second, the following principles underpin our reputation-based consensus mechanism: 1)
The liquid nature of the reputation values. The reputation value computed for a node is
based on the reputation value of the node providing the rating. 2) The temporal scoping
of reputation, so that reputation values accumulated by members in the past contribute
less to the current reputation value. 3) The openness of all reputation values to all
members of the community so that audits can be performed.
● Third, we use a side chain to store all node reputation values, eliminating the need for a
third party to manage reputation.
● Finally, we create an experimental implementation and evaluate its performance in terms
of system security and throughput.
2. REVIEW OF RELATED WORK
2.1. Consensus Algorithms
Blockchain relies heavily on consensus and it has been researched for over three decades [6], [7].
Consensus protocols have historically been used to reach an agreement on a shared among a

25
group of distributed nodes. Because designing a consensus protocol is a difficult task, it is
common to make assumptions under which the protocol is proven to function properly. These
assumptions eventually influence the characteristics of the consensus protocol. In fact, most
Blockchain Technology applications typically roll out their corresponding consensus algorithm to
fit the specific use case for the Technology [8], [9], [10].
Proof-of-Work (PoW) is by far the most widely used consensus mechanism for blockchains
introduced by Bitcoin [2], [11]. In PoW nodes acting as miners vote by the amount of computing
power they possess by trying to solve a computational challenge. The first node to solve the
challenge validates and adds a new block of transactions to the blockchain, earning a reward.
However, the generation of blocks generally requires the use of massive computational power
and introduces delay for block confirmation, resulting in low efficiency and transaction
throughput.
Proof-of-Stake (PoS) was proposed as an alternative to PoW. Nodes that want to participate in
the block creation process with PoS must prove ownership of a certain amount of coins. In
addition, they must lock a certain amount of currency, known as stake, in order to participate in
the block creation process. Delegated-Proof-of-Stake (DPoS) is a variant of PoS in which miners
are elected by other nodes. The stake of the nodes are used as the weighting parameter for votes.
2.2. Reputation Systems
The statistical estimate of a node's trustworthiness from the perspective of other nodes is known
as reputation. Its predictive power is an indicator of a node's future behavior in a network of
interacting nodes. According to the system specification, a good reputation correlates to good
behavior and a bad reputation correlates to bad behavior. Reputation is based on ratings received
with respect to a particular interaction. Nodes would typically rate other nodes on the basis of this
interaction. Specifically, a rating is a judgment from a node (origin) to another node(target) in a
scope. Reputation Systems are systems that aggregate these ratings in some way. There are three
parts to a Reputation System, the formulation, calculation and dissemination. The formulation
describes the mathematical or statistical model and a set of inputs for the assessment of reputation
values. The two main parts are: the reputation measure and the mathematical model used to
aggregate the ratings. The calculation deals with the design and implementation of the algorithm
for the assessment of reputation and finally, the dissemination is about the storage and
distribution of reputation in the network. The two popular approaches for managing
dissemination is either through a centralised authority or through decentralised nodes in a
network of nodes. Furthermore, reputation can be measured using discrete or continuous values.
Different methods have been earlier proposed to this effect.
Josang et al. [12], proposed the Beta Reputation System (BRS). In the scheme, ratings are either
positive or negative for a trustee which are thereafter considered as two events in a beta
probability distribution. The provider’s reputation is then calculated as the expected value of the
positive rating in the future derived by substituting the number of positive and negative ratings
into the beta probability density function. The scheme also uses a forgetting factor to ensure that
older ratings have a lower weight.
In the statistical approach proposed by Weng et al. [13], the credibility model is based on
statistical approaches. The scheme records the testimonies of all witnesses for each trustee and
uses that as a witness profile stored locally by every truster. The credibility of each witness is
thereafter evaluated on the basis of its history of success.

26
Other approaches include those based on fuzzy logic [14], [15], bayesian systems [16], [17], [18]
and subjective logic [19], [20].
2.3. Reputation-Based Consensus Algorithms
Reputation has been defined as a quantity derived from the underlying social network which is
globally visible to all members of the network [21], [22], [23]. Historically, reputation systems
have been known as a means of harnessing some form of reputation data. They function by
facilitating the collection, aggregation, and distribution of data about a specific entity. This data
can thereafter be used to characterize and predict that entity’s future actions [24], [25].
Essentially, users within a network can decide who they will trust and to what extent by referring
to reputation data. A reputation system, in addition to the foregoing, is a socially corrective
mechanism, as the incentive of positive reputation and the disincentive of negative reputation will
generally encourage good behaviour in the long run. After a reputation system collects reputation
data, it can be shared among users, who can then use it to evaluate other users before making
decisions about intended or future interactions, without ever having previously interacted.
Examples of the practical application of this system can be found on eCommerce websites like
Amazon or eBay where reputation attributed to a seller is influenced by ratings through previous
transactions. Another use case is in government where a country like China incentives the
behavior of the citizens through a social credit score system. Earlier works proposed possibilities
of applying a reputation-based model to distributed computing. One such was described by [26].
The downside was that the approach was not completely decentralized.
Recent studies have introduced reputation systems into the blockchain space to improve
efficiency and reliability. [27] proposed a reputation-based consensus mechanism for peer-to-peer
networks where reputation serves as the incentive for good behaviour and the node with the
highest reputation gets to publish a new block. At the end of each interaction between two nodes,
feedback is generated by the service requester and broadcast to the entire network. On reaching
the set threshold, nodes start to calculate a ranking list after which the node with the highest-
ranking reputation value publishes the new block and other nodes verify the integrity of the
newly published block. [28] also proposed a reputation-based consensus mechanism based on the
proof-of-work consensus algorithm. In their approach, a miner’s voting power is given by its
reputation. The reputation for each miner is computed based on the total amount of valid work a
miner has contributed and also the regularity of that contribution over a given period. [29]
proposed the Blockchain Reputation-Based Consensus(BRBC) mechanism in which a node in the
network must have a reputation score higher than a set threshold to be able to publish a new
block. Also, a judge is randomly selected that is responsible for updating node reputation values.
However, none of these address the behaviour of a node on a transactional basis as it interacts
with other nodes in the network.
3. BACKGROUND AND SYSTEM CONTEXT
3.1. Distributed Systems
In a distributed system, heterogeneous nodes independently run computational tasks and
communicate information via message passing. These types of systems typically consist of large
numbers of participating nodes. The behaviour of these types of systems become hard to predict
as the size of the nodes increase. A fundamental problem associated with distributed systems is
that of agreeing on a common state shared across the entire network. The reason is that large
distributed systems bring with them some form of inherent non determinism — unpredictable
events like out-of-order inputs, message delays, the sudden failure of components, or in extreme

27
cases the unethical actions of faulty or malicious nodes opposed to the goals of the system as a
whole.
Most existing agreement protocols are built on the back of some assumptions related to the
particular use case for which the protocol will be used. One of the underlying assumptions about
these protocols has to do with the network type. Dwork et al [30] categorised these as:
synchronous, partially synchronous and asynchronous. In this work, we allude to a couple of
network structure principles from [30]. Some of them are:
Definition 1: A strongly synchronous network is one in which there exists a known fixed upper
bound, δ, such that every message sent at t arrives at another point in the network at most t + δ
Definition 2: A partially synchronous network is one in which there exists a fixed upper bound, δ,
on a message’s delay and one of the following holds:
1. δ always holds, but is unpredictable.
2. δ is known, but only holds starting at some unknown time.
Furthermore, the following properties capture the essence of agreement in the network
Definition 3: If a correct node outputs a value denoted as V, then some other node proposed V.
Definition 4: If a correct node outputs a value denoted as V then all correct nodes output the value
V.
Definition 5: If all correct nodes initiated the protocol then, eventually, all correct nodes output
some value V within a finite time period t.
Definitions 3, 4 and 5 are formally referred to as Non-triviality, Agreement and Liveness
properties respectively.
3.2. Cryptographic Primitives
3.2.1. Hash Functions
We use very secure hash functions generated with the Elliptic Curve Cryptography to generate all
types of encryption needed. This method is both efficient and performant. It takes any arbitrarily
long string and converts it to a fixed-length string. The main characteristics of this hash function
is that for a given message, it is easy to compute the hash; but given the hash, it is difficult to
compute the message. Hash functions that demonstrate this property are referred to as one-way
hash functions.
3.2.2. Digital Signatures
Digital signatures enable the verification of a message. This verification ensures that a message
originated from a particular node. Digital signatures are usually a public key, denoted by pki, used
for verifying the signature of node i and a corresponding secret key, denoted by ski, used by node
i for signing a message. This process is referred to as encryption. To prevent another node or a
malicious node from impersonation, a node should not reveal this key. The purpose of the public
key is to verify a signature generated by a secret key belonging to a node in the network. A
digital signature scheme usually consists of three distinct parts:
1. a key generation algorithm,
2. a signing algorithm, and
3. a verification algorithm.

28
The key generation algorithm is used to generate a new set of key pairs. The main property
required from a digital signature scheme is that of security. Without knowledge of a secret key, it
is infeasible to find a string that passes the verification algorithm.
3.2.3. Event System
The concept of an event here is similar to traditional distributed algorithms where each event is
defined as a unit of computation across a set of nodes. An event can be described as a tuple where
pi is the node at which the event occurs, timestamp representing the exact time the event
occurred, and information about the event. This definition is based on the more general
formalism from distributed systems where each node of a distributed computation is a state
machine and the computation causes a change of state.
3.2.4. Information Propagation
Information propagation here deals with two main aspects: the first is about propagating
information from a group of nodes to all the other nodes in the network. The other aspect is the
propagation of information between two interacting nodes. The goal of information propagation
is to provide every node in the network with the most recent information. This information
propagation is the essence of communication in a distributed setting. Nodes in the network
communicate by transmitting and receiving messages. A Message is therefore a unit of
communication. A more general definition is that a message is just a finite sequence of bits. In a
distributed environment where information is constantly flowing because of the interactions,
there needs to be:
1. an efficient mechanism for distributing this information i.e publish and
2. an efficient mechanism to allow nodes to receive these specific kinds of information i.e
subscribe
Specifically, in our scheme, we use a Pub-Sub pattern as the medium for information propagation
3.2.5. Leader Election
A leader is a member of a set that is distinguished to perform some special task among other
nodes. The leader election problem is the problem of choosing a leader from a set of possible
candidates. The fundamental assumption is that every node in the network has a unique identity
and also possesses some unique characteristics which might be used as the election criteria. The
importance of electing a leader is as a result of the fact that there might be a need to conduct
some additional computation for which there is the need to have a coordinator for such
computations. It is usually a common practice that the leader is elected on the basis of some
competence with regard to the system’s operating context. Examples might be the use of compute
power as basis for election or the use of memory.
3.2.6. Voting
The goal of voting is to aggregate individual preferences among a group into a single group
ranking. Weighted voting is a voting method that affords certain nodes more privileges on the
outcome of the vote than other nodes based on some specific context for which the voting system
is deployed. Weighted voting is a form of social choice system originating from the domain of
cooperative game theory [31]. They essentially model decision making processes in which a set
of voters make decisions about the outcome of an issue. Each voter is allocated a numeric weight,

29
and the decision is carried out only if the sum of weights of voters in favour of it meets or
exceeds some specific given threshold, called the quota.
3.3. System Overview and Threat Model
We assume a strongly synchronized system where nodes are connected by a network over which
information is propagated through the exchange of messages. We first present the nodes in the
network, then we present the adversarial model. In addition, we assume that the network is
untrustworthy and unreliable, which means messages in the network can be delayed, duplicated
or lost in some cases. Furthermore, the nodes are heterogeneous and failure at a node does not
cause the failure of another node.
3.3.1. Nodes
We assume that there’s a social community that affords nodes the ability to vote about different
aspects of the system. Each node i is identified by a public key pki similar to a wallet in regular
blockchains like Bitcoin. This public key has a corresponding secret key ski with which it can use
to append its signature to transactions. Each transaction group is made up of two nodes [i,j]. Node
j gives the rating while node i is the recipient of this rating usually with respect to an interaction
between them. The transaction is said to be completed only after it is appended to the blockchain.
3.3.2. Adversary
Nodes can be in one of two broad categories: honest nodes and faulty nodes. Honest nodes are
nodes that follow the protocol precisely, and faulty nodes which do not follow the protocol.
Furthermore, a faulty node can either be the result of unresponsiveness due to a crash or lack of
information in which case they are referred to as sleepy nodes or Byzantine faulty nodes which
are nodes that act arbitrarily. We denote the total number of nodes in the network as N, the
number of faulty nodes as f which is a sum of both Byzantine nodes and sleepy nodes. We restrict
the adversary to control at most faulty nodes, where 3f + 1 ≤ N. We also assume the presence
of an adversary that can control all the faulty nodes and can coordinate their behavior.
4. METHODOLOGY
4.1. Consensus Mechanism
4.1.1. Consensus Group
In our scheme, we assume N nodes in the network, an individual node is represented as pi, i ∈ N.
Network nodes perform computation in rounds, during which a node sends messages, receives
them, and then performs some local computation on the received message [32]. Each node i is
identified by a key pair, pki which is the public key and ski is the corresponding secret key. At the
end of every interaction between nodes, rating values are generated with respect to the service. A
node will typically function in one of two ways: as the recipient node or as the rater node for the
specific interaction. When a rater node gives a rating, it broadcasts the transaction details to the
entire network. We formally denote this interaction where rater node is i and node j is the
recipient node as a tuple:

30
where pkj is the public key of the recipient node, r is the rating given by the rater node which is a
value between 0 and 1, and Eski is the encrypted transaction data signed using the rater node’s
secret key. Transactions generated between nodes for a round k are added to a list of pending
transactions waiting to be appended to the chain during the consensus phase.
At the start of every consensus round, consensus group members need to be selected and added
into a consensus group. We denote this consensus group for a round k as Gk. Members of the
consensus group are chosen from the nodes with the highest reputation values, with collective
reputation scores that exceed 50% of the total reputation values of the network. With this
approach, the size of Gk will vary depending on the reputation distribution in the network. A node
that is part of the consensus group is denoted as:
To proceed, a new leader Lk for the round k is selected. After the leader for the round k is
selected, it serves the following functions:
● Packaging all valid transactions from the list of pending transactions to a Blockk
● Calculating the new reputation values for all network nodes for Reputationk the round k
using data from transactions in the transaction list
● Broadcasting the commit message to the consensus group Gk
Algorithm 1: Node Transaction
4.1.2. Leader Selection
We use a random function to select the leader for the consensus group for each round Gk. By
doing this, there is no deterministic guarantee for which node will be selected as leader for the
consensus round k. As such, all nodes that have been selected as members of the consensus group
for the round k have equal chance of being selected. The leader’s public key is broadcasted to the
consensus group before the start of the consensus phase. The leader Lk packages all the
transactions Ti for a recent time window in a specific order and adds them into a block.
Afterwards, the leader sends a commit message to the consensus group Gk. This message contains
the newly packaged Blockk, the leader’s public key pkL, reputation list for the round k, and also a
hash of the Blockk generated using the leader’s secret key:

31
Algorithm 2: Leader Selection
4.1.3. Block Publication
After a commit message is sent by the leader Lk to the consensus group Gk, the consensus group
determines the final block to be published for the current round k. Each node pki ∈ Gk checks
the commit message sent by the leader Lk which contains both the Blockk and Hash(Blockk). First,
the node checks the pkL in the commit message to see if it matches the pkL that was broadcasted
upon the leader selection earlier; otherwise, it can ignore the commit message. It then proceeds to
check the integrity of the hash using the pkL. Afterwards, it checks the validity of the
transactions within the Blockk. If this process completes, it sends a new commit message back to
the consensus group Gk. This process continues with every node in the consensus group Gk. Upon
completion, each node pki that successfully verifies the Blockk and the ReputationList sends a
verification commit back to the consensus group Gk.
The consensus group waits until at least a certain amount of consensus members sends in this
message. This message constitutes a consensus group vote. We formalize this using a social
choice function. For a set of nodes in the consensus group for round Gk, each node has an
associated weight w assigned which is equivalent to its reputation value from the previous round
k − 1. During the consensus process, there’s a minimum quota that has to be reached for
decisions to be made. We set this quota at two-thirds of the total weight in the consensus group:
where d(Gk) represents the decision of the consensus group. Whenever the d(Gk) is 1, it means
consensus for round k has been reached.
4.2. Reputation System
The reputation of a node is determined by an analysis of the ratings it has received from others in
the past. Based on previous interactions, these ratings reflect the level of trust that other nodes
have in a specific node. In general, reputation-based systems rely on feedback to evaluate a node.
This feedback is generally in terms of the amount of satisfaction a node receives by interacting
with another node in the network. According to [33], when considering reputation information,
the source of information as well as the context must be considered. We define the reputation
principles for the approach adapted from [34] as follows:
● The liquid nature of the reputation values. The reputation value computed for a node is
based on the reputation value of the node providing the rating.

32
● The temporal scoping of reputation so that reputation values collected by members in the
past are less contributing to the current reputation value.
● The openness of all reputation values to all members in the community so that audits can
be performed.
Let si denote the reputation for a recipient node i. All nodes in the network start with a default
reputation value determined on system initialization. During node interactions, a node’s
reputation value is determined by the liquid rank algorithm [35]. This approach can be used as a
predictive metric to evaluate a node’s behaviour. For each round, a node can receive multiple
unique ratings
where the range of si is [0, 1]. Values si are then normalised as follows
We slightly modify the normalization of the rating values to prevent null values from the set of
ratings as follows
Furthermore, we define the ratings matrix S to be [sij]. After each round, these ratings will be
generated for all nodes in the network. To compute new reputation values for a node for the
round k, we blend these ratings with the rater reputation values from the previous round k − 1.
We denote this as
where = [sij] and =[rin]. rin corresponds to the rater node providing the rating.
To compute the reputation value for the round k, we then blend the initial node’s reputation value
with the current rank generated from the ratings.
where α is a constant determined on system initialization. The value is set between 0 and 1. It
determines what portion of the equation to give more priority to. If the value is set closer to 1, it
means that the newly generated reputation value will give more priority to the ratings P and less
priority to the previously generated reputation value. This is what we want as this aligns with the
reputation principles stated earlier. It helps to reduce the impact of nodes that change behaviour

33
over time. Further, to prevent reputation values from hopping, we clamp the values using a
sigmoid function as follows
Most existing reputation mechanisms rely on a central server to store, manage, and, in some
cases, distribute reputation values among network nodes. We do not need a central server in our
approach, and the reputation value for each node in the network is managed via a reputation side
chain linked to the main transaction chain. The structure of the reputation chain is such that it has
a header which contains meta information about a specific block and then the reputation values
for all the network nodes:
The ReputationListi contains a list of all network nodes with their associated reputation values for
the most recent round. This serves as a lookup data structure for future uses. The
ReputationBlocki as well as the transaction block use the standard blockchain block structure with
a hash of all the transactions for the round, a previous hash, timestamp and transactions.
A new reputation block is created along with a normal transaction block during the consensus
phase. So for a consensus round k, a Blockk which is added to the transaction chain corresponds
to a ReputationBlockk which is added to the reputation chain. As stated in section 4, part of the
duties of the consensus group is to validate the reputation calculation generated by the leader Lk.
After the consensus is reached, the leader broadcasts the new ReputationBlockk to the entire
network and as such the reputation value of all the nodes in the network is visible to all other
nodes.
5. EXPERIMENTS AND RESULTS
For our prototypes, we built an experimental protocol that implements the protocol. Thereafter,
the nodes were deployed on AWS EC2 remote server running on 16GB RAM with Amazon’s t3
processor. In the experiment, we set up 1,000 nodes. We used a default initial reputation value of
0.2. We set the value of α to 0.6, the effect of that is that we give priority to recently generated
reputation values.
We impose a round trip latency of 200ms to simulate the effect of a Wide Area Network. While
this is unlikely to occur in practice, the average of network delays across the entire network will
average out to a close enough value.
In terms of the throughput, Figure 1 shows how the throughput values change as we vary the
number of network nodes from 500 to 1,000 nodes. As the network size increases, so does the
throughput because as more nodes join the network, more messages are being transferred in the
network.

34
Figure 1: Throughput vs Number of network nodes
Also, we vary the number of transactions in a single block, we vary them between 100 and 500 to
measure the average time it takes for a new block to be produced. Figure 2 shows that it takes
more time for a new block to be produced as the transactions in a block increase. This is so
because more transactions are now in a single block and so it takes more time for those
transactions to be processed.
Figure 2: Average Block Time as the number of transactions in a single Block is varied
In our final experiment, we measure the consensus time against the block size for each block. We
observed that when the block size is relatively small around 100 transactions in each block,
consensus takes about 2 seconds. As we increase the block size, so does the consensus time
increase as well.

35
Figure 3: Consensus Time as the number of transactions in a single Block is varied
Table 1 shows the performance of our scheme against other existing consensus mechanisms.
Under certain conditions when the number of participating nodes is increased, our scheme can
achieve up to 1,100 transactions per second.
Table 1. Comparison with other consensus mechanisms
Consensus Mechanism Throughput(TPS)
Proof-of-Work 7
Proof-of-Stake 60
Proof-of-Reputation(Baseline) 800
Proof-of-Burn 854
Proof-of-Reputation 1,100
6. ANALYSIS AND SPECIFIC ATTACK VECTORS
Our consensus mechanism satisfies three important properties: safety, liveness, and fairness. The
consensus mechanism is generally safe as long as the bound on the byzantine nodes holds. The
consensus mechanism is also live. Based on the network type, a new block will eventually be
added to the chain even if there is some delay on the propagation of information. Finally, the
consensus mechanism is fair. Here, we refer to fairness in three aspects: fairness in the selection
of the consensus group members, fairness in the election of the consensus round leader and
fairness in terms of openness of all transactions and reputation values to all participants of the
network.
6.1. Safety
In the worst case scenario, we consider the existence of a byzantine node with high enough
reputation value which has successfully been included in the consensus group G for a round. The
consensus remains safe if:
1. The attacker does not control more than f nodes in the consensus group
2. The collective reputation values of the nodes compromised by the attacker is less than
1/3 of G

36
The system can continue as long as (1) or (2) do not hold. Furthermore, due to transmission
delays that may cause the system to lose some of its liveness, a node in the consensus group does
not have to wait for every other node in the consensus group but only a subset of the nodes where
the collective reputation values is greater than 2/3 of the total consensus for G.
6.2. Liveness
In a general sense, liveness guarantees that all nodes that participate in the consensus can finally
reach a consensus result. In the network assumption, there is an upper bound which represents the
limit at which a message will eventually be delivered even after it might have been delayed.
Definition 6: A protocol is live if and only if for every honest node i and a finite time t0:
1. there is a time period t1 > t0 where the node i will have received a newly committed block
2. there is a time period t2 > t0 where every other node will have received the newly
committed block
6.3. Fairness
The consensus mechanism achieves fairness in three aspects: fairness in the selection of the
consensus group members, fairness in the election of the consensus round leader and fairness in
terms of openness of all transactions and reputation values to all participants of the network.
Unlike in PoW where the consensus is centered around the compute power, the compute power
has no effect on the security in our scheme. In addition, simply having a high reputation value is
not enough because of the nature of the consensus protocol.
In most cases, a malicious node in the network exhibiting some unpredictable behaviour that
deviates from the goal of the system will receive low ratings from other nodes in the network. In
this case, the node will never even have a high enough reputation to make it into the consensus
group G therefore, it cannot successfully launch a meaningful attack that would disrupt the
security of the system.
6.4. Attack Resilience
On initialisation, the system bootstraps with all nodes having a default reputation value. At this
stage, the system is at its most vulnerable. The reason for this is that the reputation value is equal
across the entire network is equal and there is no way to distinguish between honest and
byzantine nodes especially for cases where there’s a byzantine node in the network. In addition,
reputation values need time to grow. During these early stages also, after a few rounds, the
variation in reputation values for the highest ranking nodes and the lowest ranking nodes are not
significant enough to explicitly distinguish the orientation of the nodes in the network.
Assume a consensus group with 15 nodes having a combined reputation of over 50% of the entire
network. The attacker can function in two ways in the consensus group. It may be the case that it
is:
1. It is selected as the consensus group leader L
2. It is simply a consensus group member
In the first case, as the consensus leader, the attacker node can attempt to compromise the global
shared state or global reputation by increasing its reputation which would consequently mean it
would be part of future reputation rounds or try to reduce the reputation value of other nodes or

37
both at the same time. On the basis of our random leader selection, in the presence of a malicious
node, the probability Pr(i) of that node becoming the leader is 1/15. As the size of the network
grows and consequently the size of the consensus group increases, the value of Pr(i) reduces
more and more making it more difficult for a malicious node to attempt to compromise the
system security.
Furthermore, during the voting process, the dynamic voting weight value is assigned to a node on
the basis of its reputation value. The implication of this is that the higher the reputation value, the
higher the vote weight.
For a malicious node, even if it possesses the highest reputation value in the consensus group, it
still is not able to compromise the system security as the consensus can only be reached with a
coalition of nodes with reputation value greater than 2/3 of G.
Under certain conditions as shown in the graph below, where a significant amount of consensus
nodes are controlled by a malicious node, the security can significantly come under threat. Where
the number of consensus nodes is 15, f = 4. Specifically where the combined reputation value of
the f nodes exceeds the quota, these coalition of nodes can successfully launch a successful
attack. However, as the consensus group size grows, the minimum required f nodes increases and
so a malicious node effectively has to control more nodes for which we argue that for larger
networks, this will be difficult to accomplish.
6.5. Specific Attack Vectors
6.5.1. Selfish Mining Attacks
Selfish Mining attacks [36], [37], [38] is a mining strategy where a group of miners collude to
exert power over the entire blockchain in order to increase their revenue. In selfish mining
attacks, two groups exist side-by-side: an honest group of miners following the standard protocol
and a colluding group that follows the selfish mining strategy. The selfish miners mine blocks
while keeping them secret, they continue this process until the fork created from the main chain is
longer than the main chain. In our approach, since blocks are not mined based solely on the
compute power a node possesses or a group of nodes collectively possess, this kind of attack is
impossible. Furthermore, there is no way for a node to know the nodes that will be involved in
the consensus for a round or which node will be selected as the leader for that round.
6.5.2. Eclipse Attacks
An eclipse attack [39] happens whenever a node in the network is occluded from the rest of the
network. Most of the external contact for that node is controlled by the malicious node that
launched the attack. This attack is a serious threat to any blockchain. In the case of our approach,
the effect of this type of attack is only noticeable if an attacker is able to simultaneously isolate
multiple consensus group members which is highly unlikely.
6.5.3. Flash Attacks
Flash attacks [40] happen whenever an attacker can purchase or rent compute power for a short
period with the intention of using this compute power to its advantage. This type of attack is only
feasible with network types like PoW that require the use of compute power. In our approach, an
attacker with a sufficiently large amount of compute power cannot simply launch an attack on the
basis of its compute power.

38
Table 2. Attack Resilience
Attacks PoW(Bitcoin) PBFT(ByzCoin) PoR
Liveness ✓ ╳ ✓
Flash Attacks ╳ ╳ ✓
Selfish Mining Attacks ╳ ╳ ✓
Eclipse Attacks ╳ ✓ ✓
7. CONCLUSIONS
In this work, we proposed a reputation-based consensus mechanism for distributed ledger
systems. The consensus scheme uses a social choice function where the weight of nodes that are
responsible for consensus is equivalent to the reputation value for that node. In addition, our
approach uses the liquid rank algorithm where the reputation of a node is calculated by blending
the normalized ratings by other nodes in the network for a given period with the reputation values
of the nodes giving the ratings. Finally, we built an experimental prototype to show the potential
of this approach. We remark that there are several other parts of this system which can be
improved and are left to future work.
REFERENCES
[1] M. Crosby, P. Pattanayak, S. Verma, and V. Kalyanaraman. (2016) Blockchain technology: Beyond
bitcoin. [Online]. Available: https://j2-capital.com/wp-content/uploads/2017/11/AIR- 2016-
Blockchain.pdf?forcedefault=true.
[2] A. Baliga, “Understanding blockchain consensus models,” 2017.
[3] G.-T. Nguyen and K. Kim, “A survey about consensus algorithms used in blockchain,” J. Inf.
Process. Syst., vol. 14, pp. 101–128, 2018.
[4] Quantaloop.io. (2020) Types of consensus algorithms in blockchain. [Online]. Available:
https://quantaloop.io/proof-of-work-vs-proof-of-stake-101.
[5] F. Hendrikx, K. Bubendorfer, and R. Chard, “Reputation systems: A survey and taxonomy,” Journal
of Parallel and Distributed Computing, vol. 75, pp. 184–197, 2015.
[6] V. Gramoli, “From blockchain consensus back to byzantine consensus,” Future Generation
Computer Systems, vol. 107, pp. 760–769, 2020.
[7] S. Azouvi, P. McCorry, and S. Meiklejohn, “Betting on blockchain consensus with fantomette,”
arXiv preprint arXiv:1805.06786, 2018.
[8] M. S. Ferdous, M. J. M. Chowdhury, M. A. Hoque, and A. Colman, “Blockchain consensuses
algorithms: A survey,” arXiv preprint arXiv:2001.07091, 2020.
[9] L. M. Bach, B. Mihaljevic, and M. Zagar, “Comparative analysis of blockchain consensus
algorithms,” in 2018 41st International Convention on Information and Communication Technology,
Electronics and Microelectronics (MIPRO). IEEE, 2018, pp. 1545–1550.
[10] Y. Xiao, N. Zhang, W. Lou, and Y. T. Hou, “A survey of distributed consensus protocols for
blockchain networks,” IEEE Communications Surveys & Tutorials, vol. 22, no. 2, pp. 1432–1465,
2020.
[11] A. Gervais, G. O. Karame, K. Wu ̈st, V. Glykantzis, H. Ritzdorf, and S. Capkun, “On the security and
performance of proof of work blockchains,” in Proceedings of the 2016 ACM SIGSAC conference on
computer and communications security, 2016, pp. 3–16.
[12] A. Josang and R. Ismail, “The beta reputation system,” in Proceedings of the 15th bled electronic
commerce conference, vol. 5, 2002, pp. 2502– 2511.
[13] J. Weng, Z. Shen, C. Miao, A. Goh, and C. Leung, “Credibility: How agents can handle unfair third-
party testimonies in computational trust models,” IEEE Transactions on Knowledge and Data
Engineering, vol. 22, no. 9, pp. 1286–1298, 2009.

39
[14] S. Song, K. Hwang, R. Zhou, and Y.-K. Kwok,“Trusted p2p transactions with fuzzy reputation
aggregation,” IEEE Internet computing, vol. 9, no. 6, pp. 24–34, 2005.
[15] K.K.Bharadwaj and M.Y.H.Al-Shamri,“Fuzzy computational models for trust and reputation
systems,” Electronic commerce research and applications, vol. 8, no. 1, pp. 37–47, 2009.
[16] W. L. Teacy, M. Luck, A. Rogers, and N. R. Jennings, “An efficient and versatile approach to trust
and reputation using hierarchical bayesian modelling,” Artificial Intelligence, vol. 193, pp. 149–185,
2012.
[17] M. Tavakolifard and S. J. Knapskog, “A probabilistic reputation algorithm for decentralized multi-
agent environments,” Electronic Notes in Theoretical Computer Science, vol. 244, pp. 139–149,
2009.
[18] A. Whitby, A. Jøsang, and J. Indulska, “Filtering out unfair ratings in bayesian reputation systems,”
in Proc. 7th Int. Workshop on Trust in Agent Societies, vol. 6. Citeseer, 2004, pp. 106–117.
[19] A. Jøsang, Subjective logic. Springer, 2016.
[20] A. Jøsang and T. Bhuiyan, “Optimal trust network analysis with subjective logic,” in 2008 Second
International Conference on Emerging Security Information, Systems and Technologies. IEEE, 2008,
pp. 179– 184.
[21] L. C. Freeman, “Centrality in social networks conceptual clarification,” Social networks, vol. 1, no. 3,
pp. 215–239, 1978.
[22] L. Kleinrock, R. Ostrovsky, and V. Zikas, “Proof-of-reputation blockchain with nakamoto fallback,”
in International Conference on Cryptology in India. Springer, 2020, pp. 16–38.
[23] J. Horton and J. Golden,“Reputation Inflation An Online Marketplace,” New York I, vol. 1, 2015.
[24] G. Swamynathan, K. C. Almeroth, and B. Y. Zhao, “The design of a reliable reputation system,”
Electronic Commerce Research, vol. 10, no. 3, pp. 239–270, 2010.
[25] K. Hoffman, D. Zage, and C. Nita-Rotaru, “A survey of attack and defense techniques for reputation
systems,” ACM Computing Surveys (CSUR), vol. 42, no. 1, pp. 1–31, 2009.
[26] M. Gupta, P. Judge, and M. Ammar, “A reputation system for peer-to-peer networks,” in Proceedings
of the 13th international workshop on Network and operating systems support for digital audio and
video, 2003, pp. 144–152.
[27] F. Gai, B. Wang, W. Deng, and W. Peng, “Proof of reputation: A reputation-based consensus
protocol for peer-to-peer network,” in International Conference on Database Systems for Advanced
Applications. Springer, 2018, pp. 666–681.
[28] J. Yu, D. Kozhaya, J. Decouchant, and P. Esteves-Verissimo,“Repucoin: Your reputation is your
power,” IEEE Transactions on Computers, vol. 68, no. 8, pp. 1225–1237, 2019.
[29] M. T. de Oliveira, L. H. Reis, D. S. Medeiros, R. C. Carrano, S. D. Olabarriaga, and D. M. Mattos,
“Blockchain reputation-based consensus: A scalable and resilient mechanism for distributed
mistrusting applications,” Computer Networks, vol. 179, p. 107367, 2020.
[30] C. Dwork, N. Lynch, and L. Stockmeyer, “Consensus in the presence of partial synchrony,” Journal
of the ACM (JACM), vol. 35, no. 2, pp. 288–323, 1988.
[31] G. Chalkiadakis, E. Elkind, and M. Wooldridge, “Computational aspects of cooperative game
theory,” Synthesis Lectures on Artificial Intelligence and Machine Learning, vol. 5, no. 6, pp. 1–168,
2011.
[32] P. Berman, J. A. Garay, K. J. Perry et al., “Towards optimal distributed consensus,” in FOCS, vol. 89.
Citeseer, 1989, pp. 410–415.
[33] L. Xiong and L. Liu, “Peertrust: Supporting reputation-based trust for peer-to-peer electronic
communities,” IEEE transactions on Knowledge and Data Engineering, vol. 16, no. 7, pp. 843–857,
2004.
[34] A. Kolonin and S. SingularityNET, “Reputation systems for human-computer environments,”
Complexity, Informatics and Cybernetics, 2019.
[35] A. Kolonin, B. Goertzel, D. Duong, and M. Ikle, “A reputation system for artificial societies,” arXiv
preprint arXiv:1806.07342, 2018.
[36] I. Eyal and E. G. Sirer, “Majority is not enough: Bitcoin mining is vulnerable,” in International
conference on financial cryptography and data security. Springer, 2014, pp. 436–454.
[37] C. Grunspan and R. Perez-Marco, “On profitability of selfish mining,” arXiv preprint
arXiv:1805.08281, 2018.
[38] K. A. Negy, P. R. Rizun, and E. G. Sirer, “Selfish mining re-examined,” in International Conference
on Financial Cryptography and Data Security. Springer, 2020, pp. 61–78.

40
[39] E. Heilman, A. Kendler, A. Zohar, and S. Goldberg, “Eclipse attacks on bitcoin’s peer-to-peer
network,” in 24th {USENIX} Security Symposium ({USENIX} Security 15), 2015, pp. 129–144.
[40] J. Bonneau, "Why Buy When You Can Rent?" in International Conference on Financial
Cryptography and Data Security. Springer, 2016, pp. 19–26.
AUTHORS
Oladotun Aluko received his BSc(2017) in Computer Science and Engineering from
Obafemi Awolowo University, Nigeria. He is currently working on his MSc in Big Data
Analytics and Artificial Intelligence at Novosibirsk State University, Novosibirsk,
Russia. His research interests are in Distributed Computing, Blockchain Technology,
Machine Learning and Cloud Databases.
Anton Kolonin received his PhD in 1998 after he independently developed a software-
algorithmic complex for processing geophysical data, introduced into production in many
CIS countries. He has also participated as a leader or lead architect in many projects to
develop algorithms and software, including those related to the use of AI, including the
recognition of static text, moving objects, music, extracting information from texts and
identifying events on financial markets – in Russian and foreign companies. Since 2017,
he has also been a software architect for AI and blockchain in the Singularity NET
project, leading projects on unsupervised language learning and reputation systems.

PROOF-OF-REPUTATION: AN ALTERNATIVE CONSENSUS MECHANISM FOR BLOCKCHAIN SYSTEMS

More Related Content

What's hot (18)

Similar to PROOF-OF-REPUTATION: AN ALTERNATIVE CONSENSUS MECHANISM FOR BLOCKCHAIN SYSTEMS (20)

Recently uploaded (20)

PROOF-OF-REPUTATION: AN ALTERNATIVE CONSENSUS MECHANISM FOR BLOCKCHAIN SYSTEMS