PoS, DPoS, PBFT...Innovative consensus emerges one after another. Is PoW outdated?

In today's blockchain world full of various innovative consensuses (PoS, DPoS, PBFT), is PoW outdated? What is the future of PoW? Is it necessary to do some consensus optimization and innovation based on PoW?

We believe that PoW is still the core or foundation of other consensuses and has an irreplaceable position. Why do we say that? Let's try to see through the connection between these consensuses.

Distributed consensus

Looking back at history, Satoshi Nakamoto's PoW was not a continuation of the academic research on distributed consensus. He initially did not know whether the solution he proposed corresponded to a certain academic topic. Only later did everyone think that Satoshi Nakamoto's PoW roughly provided a solution for distributed consensus in the Byzantine environment. Only then did we have an academic perspective on PoW and could connect the issues of PoW research with academic research on distributed consensus.

When it comes to distributed consensus, we have to mention the "FLP Impossibility" theorem: In a minimal asynchronous model system where the network is reliable but node failures are allowed (even if only one) there is no deterministic consensus algorithm that can solve the consistency problem. The paper "Impossibility of Distributed Consensus with One Faulty Process" that proposed and proved this theorem was published in 1985 by three scientists, Fischer, Lynch and Patterson. The revelation of this theorem is that the distributed consensus that is feasible in engineering now must have compromised in some aspects.

In fact, there are currently two directions of compromise: one is to compromise on asynchrony, and the other is to compromise on determinism.

The so-called compromise with asynchrony is to set a timeout mechanism, assuming that messages will not be delayed indefinitely. The DLS algorithm ("Consensus in the Presence of Partial Synchrony") and the PBFT algorithm (Practical Byzantine Fault Tolerance) are both solutions in this direction. Furthermore, the DLS paper reveals the two goals of asynchronous consensus in a Byzantine environment: security (consistency) and liveness (availability). Both of these solutions make synchronization assumptions for liveness, not for security. That is, if the real environment does not meet the synchronization assumption, it will only cause availability problems, not consistency problems. This is why we often say that PBFT is a consensus algorithm that prioritizes consistency over availability.

The so-called compromise on certainty is the new world introduced by Nakamoto, where certainty is probabilistic. In this direction, a probabilistically secure consensus scheme is jointly constructed through proof of work, block rewards, and peer-to-peer networks. It seems that in terms of the layout, the Nakamoto consensus is just one of the compromise solutions to the asynchronous Byzantine consensus problem defined by consensus researchers, but the introduction of probabilistic certainty not only solves the traditional asynchronous Byzantine problem, but also solves a bigger problem: any number of nodes can participate in the system in an open way, and no participant has to know the complete set of participants. This is very remarkable, and this is exactly the environment that the public chain system really faces. In practice, the Nakamoto consensus that compromises certainty has achieved great success.

In fact, the innovative public chain consensus mechanism is nothing more than these two compromise directions. Some representatives of them are introduced below.

Compromise Analysis of Different Consensuses

PoS

Pos is essentially PoW, which is a compromise on certainty. Staking can be regarded as the entry threshold for participation. By raising the threshold, the size of the participant set is reduced. A smaller participant set has two benefits: 1. The total workload of the entire system will be reduced, and energy consumption will be reduced; 2. The asynchronous communication network is smaller in size, and the possibility of partition caused by delay is smaller, and a smaller block interval can be set. This is why PoS claims that it has a higher TPS and is more energy-efficient.

DPoS+PBFT

DPoS+PBFT is of course PBFT in essence, which is a compromise for asynchrony. The purpose of DPoS is to select a set of participants from open public chain participants to which the PBFT algorithm can be applied. This set meets three conditions: 1. The scale is small enough, otherwise the communication volume is huge; 2. The total number of sets is determined, so that it is an asynchronous consensus problem that PBFT can solve; 3. The number of malicious participants (Byzantine nodes) is less than 1/3. The Staking mechanism plays two roles in this: 1. It serves as a threshold to screen users; 2. It serves as an incentive mechanism to reduce the proportion of malicious nodes.

Tetris

Tetris is a consensus algorithm based on knowledge reasoning pioneered by YeeCo. It is essentially BFT, which is a compromise for asynchrony. Tetris itself is a solution to the standard Byzantine problem, focusing on its high performance and proven security. Its BFT participant set is selected through a pluggable upper-layer protocol, including PoW, DPoS, VRF, etc.

Algorand

Algorand is essentially BFT, a compromise with asynchrony. Algorand's method of selecting a set of participants to which the BFT algorithm can be applied is very special and clever, using VRF (verifiable random function), similar to each participant carrying a lottery device, so that they can know whether they are selected without communicating with other participants. It seems to be very energy-efficient, but there are several problems to be solved: 1. It is necessary to set a threshold for participants, because the cost of creating a private key that can participate in the network is very low. In fact, Algorand gives weights according to the balance of participants; 2. The lottery needs to set a winning rate, and the total number of participants needs to be known, which is itself a consensus problem. Algorand rewards nodes online and tries to achieve a situation where the online balance is equivalent to the total balance of the system, so that the number of participants can be inferred. This is actually a Byzantine fault-tolerant system that requires the entire network to be online and honest with a balance > 2/3.

DAG

DAG (directed acyclic graph), whether it is IOTA with transactions as graph vertices or Conflux with blocks as graph vertices, still works on PoW, which is a compromise on certainty. If the previous consensus was to avoid forks, then DAG is to control forks at a certain level, thereby improving the system's TPS. The essence of DAG is still proof of work, but the longest chain principle has been changed to the most difficult graph structure principle.

PoW is the core or foundation

We roughly summarize the consensus model as shown below:

The difference in consensus mechanisms is actually the difference in the paths from "open set of network participants" to "consensus". For example, going directly from "open set of network participants" to "consensus" is PoW; going to "open set of accounting participants" first and then to "consensus" is PoS; going to "closed set of accounting participants" first and then to "consensus" is X+BFT. X depends on what tools the "closed set of accounting participants" uses, which can be DPoS or VRF.

It can be seen that so many consensus algorithms do not actually invent much new things. As long as the consensus is formed from an "open set of participants" (nodes can enter and exit freely), this model can only be adopted: nodes are incentivized to continue to establish consensus results based on a consensus result with probabilistic certainty, which is the Nakamoto consensus.

The significance of PoW for the current innovative consensus lies in two points:

1. If consensus involves forming a consensus from an "open set of participants" (nodes can enter and exit freely), it is essentially still the Nakamoto consensus, which is the "kernel" in the previous statement "PoW is still the core or foundation of other consensuses at present";

2. If the consensus involves changing the "open participant set" to a "closed participant set", then the Staking mechanism is still needed. Where does the consensus on the value of Staking come from? It is still transmitted from the PoW chain. This is the "foundation" in the previous statement that "PoW is still the core or foundation of other consensuses at present";

The essential advantages of PoW

PoW supporters have given many qualitative advantages of PoW, such as naturalness, credibility, and purity. As for energy consumption, some people think it is a disadvantage of PoW, while others think that cost is the cornerstone of security.

Let me talk about the essential advantages of PoW in my opinion:

The Ultimate Openness of Consensus

PoW is designed for an open set of participants. This openness (instability of the participant set) is constantly challenging the consensus, whether it is the earliest cold start, or the emergence of mining pools, ASIC mining machines, fluctuations in coin prices, and fluctuations in difficulty.

PoW adopts an extremely simple mechanism to solve the openness problem . PoW does not need to apply any patches or do any governance for any scenario or any stage. It always relies on its core mechanism to drive the entire system and has withstood the test of time. In our opinion, some systems grasp the essence, are clever and simple, and have strong adaptability, while some systems are full of compensation mechanisms and are constantly changing with the changes in the environment. PoW is an excellent representative of the former.

PoW is a permissionless system with a bottom line . At any time, computing power can enter the market and compete with existing computing power at the same starting line. Staking investment and computing power investment look very similar, but in a game environment, the difference in starting point parameters will lead to completely different patterns. If the blockchain system is a system that is prone to the "Matthew effect", the external entity computing power investment of PoW is an important parameter to resist this effect. On the one hand, the lagging adjustment mechanism increases the difficulty of the system entering a steady state; on the other hand, the external entity computing power has better independence than Staking. The positive feedback of the consensus advantage on the business advantage in PoW is one-way, while the positive feedback of the consensus advantage on the business advantage in Staking is two-way.

The Ultimate in Consensus Security

The security issues that are often discussed in consensus, such as 51% attacks and more than 1/3 of Byzantine nodes, are all under the consensus rules. The blockchain system is ultimately a system running on a physical network. Compared with malicious acts under the consensus rules, attacks at the physical network level are a dimensionality reduction attack on the blockchain system.

Pure PoW believes that the weight of a node should not be determined by virtual resources (Staking), but by physical resources (computing power).

The former can easily fall into a situation where virtual resources are continuously invested, but there is not enough driving force for the growth of network node scale. That is, the economic scale of the system and its ability to resist threats from the network are not matched, and there is no endogenous mechanism to promote the match.

The latter's resource investment actually corresponds to the investment in network discourse power. The initial computing power investment is the investment in running more nodes. Later, professional mining machines and mining pools appeared, and the situation of incomplete matching between computing power and nodes appeared. However, since those who invest in computing power have invested, they will definitely take measures to obtain the corresponding network discourse power. The economic scale of the system and its ability to resist threats from the network are matched. The more virtual resources are considered in the node weight, the greater the difference between the node's discourse power in the system and its discourse power in the physical network. Without considering the PoW of virtual resources, the node's discourse power in the system is consistent with its discourse power in the physical network. The security of PoW is reflected in this. The root of the cost support for security lies in the physical network environment.

Do "consensus openness" and "consensus security" look familiar to you? They are exactly the two corners of the "blockchain impossible triangle" (that is, it is impossible to achieve scalability, decentralization, and security at the same time, only two of the three can be achieved).

The essential advantage of PoW is that it truly provides a highly decentralized and highly secure consensus mechanism. And scalability is exactly the direction in which PoW consensus needs to develop and innovate.

Author: YeeCo, authorized to reprint by Babbitt Information