News
An enhanced consensus algorithm for blockchain
The introduction of the link and reputation evaluation concepts aims to improve the stability and security of the consensus mechanism, decrease the likelihood of malicious nodes joining the consensus, and increase the reliability of the selected consensus nodes.
The link model structure based on joint action
Through the LINK between nodes, all the LINK nodes engage in consistent activities during the operation of the consensus mechanism. The reputation evaluation mechanism evaluates the trustworthiness of nodes based on their historical activity status throughout the entire blockchain. The essence of LINK is to drive inactive nodes to participate in system activities through active nodes. During the stage of selecting leader nodes, nodes are selected through self-recommendation, and the reputation evaluation of candidate nodes and their LINK nodes must be qualified. The top 5 nodes of the total nodes are elected as leader nodes through voting, and the nodes in their LINK status are candidate nodes. In the event that the leader node goes down, the responsibility of the leader node is transferred to the nodes in its LINK through the view-change. The LINK connection algorithm used in this study is shown in Table 2, where LINKm is the linked group and LINKP is the percentage of linked nodes.
Table 2 LINK connection algorithm.
Node type
This paper presents a classification of nodes in a blockchain system based on their functionalities. The nodes are divided into three categories: leader nodes (LNs), follower nodes (FNs), and general nodes (Ns). The leader nodes (LNs) are responsible for producing blocks and are elected through voting by general nodes. The follower nodes (FNs) are nodes that are linked to leader nodes (LNs) through the LINK mechanism and are responsible for validating blocks. General nodes (N) have the ability to broadcast and disseminate information, participate in elections, and vote. The primary purpose of the LINK mechanism is to act in combination. When nodes are in the LINK, there is a distinction between the master and slave nodes, and there is a limit to the number of nodes in the LINK group (NP = {n1, nf1, nf2 ……,nfn}). As the largest proportion of nodes in the system, general nodes (N) have the right to vote and be elected. In contrast, leader nodes (LNs) and follower nodes (FNs) do not possess this right. This rule reduces the likelihood of a single node dominating the block. When the system needs to change its fundamental settings due to an increase in the number of nodes or transaction volume, a specific number of current leader nodes and candidate nodes need to vote for a reset. Subsequently, general nodes need to vote to confirm this. When both confirmations are successful, the new basic settings are used in the next cycle of the system process. This dual confirmation setting ensures the fairness of the blockchain to a considerable extent. It also ensures that the majority holds the ultimate decision-making power, thereby avoiding the phenomenon of a small number of nodes completely controlling the system.
After the completion of a governance cycle, the blockchain network will conduct a fresh election for the leader and follower nodes. As only general nodes possess the privilege to participate in the election process, the previous consortium of leader and follower nodes will lose their authorization. In the current cycle, they will solely retain broadcasting and receiving permissions for block information, while their corresponding incentives will also decrease. A diagram illustrating the node status can be found in Fig. 1.
Election method
The election method adopts the node self-nomination mode. If a node wants to participate in an election, it must form a node group with one master and three slaves. One master node group and three slave node groups are inferred based on experience in this paper; these groups can balance efficiency and security and are suitable for other project collaborations. The successfully elected node joins the leader node set, and its slave nodes enter the follower node set. Considering the network situation, the maximum threshold for producing a block is set to 1 s. If the block fails to be successfully generated within the specified time, it is regarded as a disconnected state, and its reputation score is deducted. The node is skipped, and in severe cases, a view transformation is performed, switching from the master node to the slave node and inheriting its leader’s rights in the next round of block generation. Although the nodes that become leaders are high-reputation nodes, they still have the possibility of misconduct. If a node engages in misconduct, its activity will be immediately stopped, its comprehensive reputation score will be lowered, it will be disqualified from participating in the next election, and its equity will be reduced by 30%. The election process is shown in Fig. 2.
Figure 2
Incentives and penalties
To balance the rewards between leader nodes and ordinary nodes and prevent a large income gap, two incentive/penalty methods will be employed. First, as the number of network nodes and transaction volume increase, more active nodes with significant stakes emerge. After a prolonged period of running the blockchain, there will inevitably be significant class distinctions, and ordinary nodes will not be able to win in the election without special circumstances. To address this issue, this paper proposes that rewards be reduced for nodes with stakes exceeding a certain threshold, with the reduction rate increasing linearly until it reaches zero. Second, in the event that a leader or follower node violates the consensus process, such as by producing a block out of order or being unresponsive for an extended period, penalties will be imposed. The violation handling process is illustrated in Fig. 3.
Figure 3
Violation handling process.
Comprehensive reputation evaluation and election mechanism based on historical transactions
This paper reveals that the core of the DPoS consensus mechanism is the election process. If a blockchain is to run stably for a long time, it is essential to consider a reasonable election method. This paper proposes a comprehensive reputation evaluation election mechanism based on historical records. The mechanism considers the performance indicators of nodes in three dimensions: production rate, tokens, and validity. Additionally, their historical records are considered, particularly whether or not the nodes have engaged in malicious behavior. For example, nodes that have ever been malicious will receive low scores during the election process unless their overall quality is exceptionally high and they have considerable support from other nodes. Only in this case can such a node be eligible for election or become a leader node. The comprehensive reputation score is the node’s self-evaluation score, and the committee size does not affect the computational complexity.
Moreover, the comprehensive reputation evaluation proposed in this paper not only is a threshold required for node election but also converts the evaluation into corresponding votes based on the number of voters. Therefore, the election is related not only to the benefits obtained by the node but also to its comprehensive evaluation and the number of voters. If two nodes receive the same vote, the node with a higher comprehensive reputation is given priority in the ranking. For example, in an election where node A and node B each receive 1000 votes, node A’s number of stake votes is 800, its comprehensive reputation score is 50, and only four nodes vote for it. Node B’s number of stake votes is 600, its comprehensive reputation score is 80, and it receives votes from five nodes. In this situation, if only one leader node position remains, B will be selected as the leader node. Displayed in descending order of priority as comprehensive credit rating, number of voters, and stake votes, this approach aims to solve the problem of node misconduct at its root by democratizing the process and subjecting leader nodes to constraints, thereby safeguarding the fundamental interests of the vast majority of nodes.
Comprehensive reputation evaluation
This paper argues that the election process of the DPoS consensus mechanism is too simplistic, as it considers only the number of election votes that a node receives. This approach fails to comprehensively reflect the node’s actual capabilities and does not consider the voters’ election preferences. As a result, nodes with a significant stake often win and become leader nodes. To address this issue, the comprehensive reputation evaluation score is normalized considering various attributes of the nodes. The scoring results are shown in Table 3.
Table 3 Comprehensive reputation evaluation.
Since some of the evaluation indicators in Table 3 are continuous while others are discrete, different normalization methods need to be employed to obtain corresponding scores for different indicators. The continuous indicators include the number of transactions/people, wealth balance, network latency, network jitter, and network bandwidth, while the discrete indicators include the number of violations, the number of successful elections, and the number of votes. The value range of the indicator “number of transactions/people” is (0,1), and the value range of the other indicators is (0, + ∞). The equation for calculating the “number of transactions/people” is set as shown in Eq. (1).
$$A_{1} = \left\{ {\begin{array}{*{20}l} {0,} \hfill & {{\text{G}} = 0} \hfill \\ {\frac{{\text{N}}}{{\text{G}}}*10,} \hfill & {{\text{G}} > 0} \hfill \\ \end{array} } \right.$$
(1)
where N represents the number of transactional nodes and G represents the number of transactions. It reflects the degree of connection between the node and other nodes. Generally, nodes that transact with many others are safer than those with a large number of transactions with only a few nodes. The limit value of each item, denoted by x, is determined based on the situation and falls within the specified range, as shown in Eq. (2). The wealth balance and network bandwidth indicators use the same function to set their respective values.
$${A}_{i}=20*\left(\frac{1}{1+{e}^{-{a}_{i}x}}-0.5\right)$$
(2)
where x indicates the value of this item and expresses the limit value.
In Eq. (3), x represents the limited value of this indicator. The lower the network latency and network jitter are, the higher the score will be.
The last indicators, which are the number of violations, the number of elections, and the number of votes, are discrete values and are assigned different scores according to their respective ranges. The scores corresponding to each count are shown in Table 4.
$$A_{3} = \left\{ {\begin{array}{*{20}l} {10*\cos \frac{\pi }{200}x,} \hfill & {0 \le x \le 100} \hfill \\ {0,} \hfill & {x > 100} \hfill \\ \end{array} } \right.$$
(3)
Table 4 Score conversion.
The reputation evaluation mechanism proposed in this paper comprehensively considers three aspects of nodes, wealth level, node performance, and stability, to calculate their scores. Moreover, the scores obtain the present data based on historical records. Each node is set as an M × N dimensional matrix, where M represents M times the reputation evaluation score and N represents N dimensions of reputation evaluation (M < = N), as shown in Eq. (4).
$${\text{N}} = \left( {\begin{array}{*{20}c} {a_{11} } & \cdots & {a_{1n} } \\ \vdots & \ddots & \vdots \\ {a_{m1} } & \cdots & {a_{mn} } \\ \end{array} } \right)$$
(4)
The comprehensive reputation rating is a combined concept related to three dimensions. The rating is set after rating each aspect of the node. The weight w and the matrix l are not fixed. They are also transformed into matrix states as the position of the node in the system changes. The result of the rating is set as the output using Eq. (5).
$$\text{T}=\text{lN}{w}^{T}=\left({l}_{1}\dots {\text{l}}_{\text{m}}\right)\left(\begin{array}{ccc}{a}_{11}& \cdots & {a}_{1n}\\ \vdots & \ddots & \vdots \\ {a}_{m1}& \cdots & {a}_{mn}\end{array}\right){\left({w}_{1}\dots {w}_{n}\right)}^{T}$$
(5)
Here, T represents the comprehensive reputation score, and l and w represent the correlation coefficient. Because l is a matrix of order 1*M, M is the number of times in historical records, and M < = N is set, the number of dimensions of l is uncertain. Set the term l above to add up to 1, which is l1 + l2 + …… + ln = 1; w is also a one-dimensional matrix whose dimension is N*1, and its purpose is to act as a weight; within a certain period of time, w is a fixed matrix, and w will not change until the system changes the basic settings.
Assume that a node conducts its first comprehensive reputation rating, with no previous transaction volume, violations, elections or vote. The initial wealth of the node is 10, the latency is 50 ms, the jitter is 100 ms, and the network bandwidth is 100 M. According to the equation, the node’s comprehensive reputation rating is 41.55. This score is relatively good at the beginning and gradually increases as the patient participates in system activities continuously.
Voting calculation method
To ensure the security and stability of the blockchain system, this paper combines the comprehensive reputation score with voting and randomly sorts the blocks, as shown in Eqs. (3–6).
$$Z=\sum_{i=1}^{n}{X}_{i}+nT$$
(6)
where Z represents the final election score, Xi represents the voting rights earned by the node, n is the number of nodes that vote for this node, and T is the comprehensive reputation score.
The voting process is divided into stake votes and reputation votes. The more reputation scores and voters there are, the more total votes that are obtained. In the early stages of blockchain operation, nodes have relatively few stakes, so the impact of reputation votes is greater than that of equity votes. This is aimed at selecting the most suitable node as the leader node in the early stage. As an operation progresses, the role of equity votes becomes increasingly important, and corresponding mechanisms need to be established to regulate it. The election vote algorithm used in this paper is shown in Table 5.
Table 5 Election vote counting algorithm.
This paper argues that the election process utilized by the original DPoS consensus mechanism is overly simplistic, as it relies solely on the vote count to select the node that will oversee the entire blockchain. This approach cannot ensure the security and stability of the voting process, and if a malicious node behaves improperly during an election, it can pose a significant threat to the stability and security of the system as well as the safety of other nodes’ assets. Therefore, this paper proposes a different approach to the election process of the DPoS consensus mechanism by increasing the complexity of the process. We set up a threshold and optimized the vote-counting process to enhance the security and stability of the election. The specific performance of the proposed method was verified through experiments.
The election cycle in this paper can be customized, but it requires the agreement of the blockchain committee and general nodes. The election cycle includes four steps: node self-recommendation, calculating the comprehensive reputation score, voting, and replacing the new leader. Election is conducted only among general nodes without affecting the production or verification processes of leader nodes or follower nodes. Nodes start voting for preferred nodes. If they have no preference, they can use the LINK mechanism to collaborate with other nodes and gain additional rewards.
View changes
During the consensus process, conducting a large number of updates is not in line with the system’s interests, as the leader node (LN) and follower node (FN) on each node have already been established. Therefore, it is crucial to handle problematic nodes accurately when issues arise with either the LN or FN. For instance, when a node fails to perform its duties for an extended period or frequently fails to produce or verify blocks within the specified time range due to latency, the system will precisely handle them. For leader nodes, if they engage in malicious behavior such as producing blocks out of order, the behavior is recorded, and their identity as a leader node is downgraded to a follower node. The follower node inherits the leader node’s position, and the nature of their work is transformed as they swap their responsibilities of producing and verifying blocks with their original work. This type of behavior will not significantly affect the operation of the blockchain system. Instead of waiting until the end of the current committee round to punish malicious nodes, dynamic punishment is imposed on the nodes that affect the operation of the blockchain system to maintain system security. The view change operation is illustrated in Fig. 4.
Figure 4
In traditional PBFT, view changes are performed according to the view change protocol by changing the view number V to the next view number V + 1. During this process, nodes only receive view change messages and no other messages from other nodes. In this paper, the leader node group (LN) and follower node group (FN) are selected through an election of the LINK group. The node with LINKi[0] is added to the LN leader node group, while the other three LINK groups’ follower nodes join the FN follower node group since it is a configuration pattern of one master and three slaves. The view change in this paper requires only rearranging the node order within the LINK group to easily remove malicious nodes. Afterward, the change is broadcast to other committee nodes, and during the view transition, the LINK group does not receive block production or verification commands from the committee for stability reasons until the transition is completed.