← Back to list
Security & Privacy
#블록체인#합의 알고리즘#작업증명#지분증명#탈중앙화
Last updated · 2026-11-08

Blockchain Consensus Mechanisms (PoW, PoS, DPoS)

1. Overview

A. Definition

A Blockchain Consensus Mechanism is the set of rules that makes many mutually distrusting, anonymous nodes converge on one identical state for the next block of the ledger and its ordering, without any central trusted authority. Its essence is twofold: ① in an environment where anyone can join anonymously, it uses cost to deter the Sybil Attack, in which one participant endlessly clones identities to monopolize voting power, and ② it combines a fork choice rule that selects a single canonical chain even when forks occur.

Blockchain consensus shares its roots with classical distributed consensus but rests on different premises. Paxos/Raft ([[distributed-consensus]]) and PBFT ([[byzantine-fault-tolerance]]) assume a permissioned environment where the set of participating nodes is fixed in advance, so "who votes" is already settled. By contrast, in the permissionless world that Bitcoin opened, nodes come and go freely and have no identity at all. If you naively applied "one node, one vote," an attacker could clone tens of thousands of nodes to seize a majority; hence the idea of binding voting power to a scarce resource (computing power, stake) so that cloning becomes economically impossible. This idea is precisely the Nakamoto Consensus proposed by Satoshi Nakamoto in 2008, which has since branched into proof of stake, delegated proof of stake, and more, forming the heart of today's public blockchains.

The standing of consensus in a blockchain is more than a mere "synchronization technology." Every value proposition of a blockchain—immutability, transparency, decentralization—ultimately rests on consensus's guarantee that "no one can arbitrarily alter the ledger." If consensus breaks (e.g., a 51% takeover), double spending, censorship, and history rewriting become possible, and the trust machine that is the blockchain collapses. Consensus is therefore the single most fundamental design decision that simultaneously defines a blockchain's security, performance, and decentralization, and which consensus one chooses determines what that network protects and what it gives up.

B. Background and the Requirements Consensus Must Satisfy

Electronic cash must absolutely prevent double spending—"spending the same money twice." Traditionally a central ledger such as a bank or card company guaranteed "this account has only one balance," but in P2P money with no central authority, who finalizes the order of transactions was the hard problem. When many nodes receive transactions simultaneously they may see different orders, and a malicious user can then broadcast two conflicting transactions to different nodes and spend the money on both sides. Blockchain consensus is the device that makes "the whole world agree on the global total order of transactions" and thereby blocks double spending at the source.

For this consensus to be practical it must satisfy four requirements at once. First, Safety—honest nodes do not finalize mutually contradictory ledgers. Second, Liveness—a legitimate transaction is eventually included in a block. Third, Sybil Resistance—one cannot increase influence for free by cloning identities. Fourth, Decentralization—a minority cannot censor or monopolize. Real designs endlessly trade off among these requirements versus throughput and confirmation latency, and the blockchain trilemma formalized by Vitalik Buterin (it is hard to maximize decentralization, security, and scalability simultaneously) is the underlying sketch for every consensus design.

2. Classification and Overall Structure of Consensus Algorithms

Blockchain consensus splits along two axes: "what voting power is bound to (the Sybil-resistance resource)" and "whether confirmation is probabilistic or immediate (finality)." The structure diagram below shows the major families and their branches.

flowchart TB
    ROOT["Blockchain consensus algorithms"]
    ROOT --> COSTLY["Resource-investment type (permissionless)"]
    ROOT --> VOTE["Voting-quorum type (permissioned/semi)"]
    COSTLY --> POW["Proof of Work (PoW)<br/>resource = compute"]
    COSTLY --> POS["Proof of Stake (PoS)<br/>resource = staked stake"]
    POS --> DPOS["Delegated PoS (DPoS)<br/>elect few delegates"]
    POS --> LPOS["Nominated/Liquid PoS<br/>(NPoS, LPoS)"]
    VOTE --> BFTV["BFT family (PBFT, Tendermint)<br/>instant finality"]
    VOTE --> POA["Proof of Authority (PoA)<br/>trusted validators"]
    POW -.->|"probabilistic finality"| FIN1["longest-chain rule"]
    BFTV -.->|"absolute finality"| FIN2["quorum signatures"]

A. Proof of Work (PoW) — Nakamoto Consensus

Proof of Work (PoW) grants the right to propose a block to the node that first solves a "hard to solve, easy to verify" computational puzzle. In Bitcoin, a miner changes the block header's nonce repeatedly until the SHA-256 hash falls below a target value. Because a hash cannot be inverted ([[cryptographic-hash-function]]), only brute force yields an answer, and the probability of success is proportional to the hash power invested. In other words, the value of "one vote" is bound to the real external cost of electricity and hardware, so no matter how many identities you clone, you gain no influence without computing power. This is PoW's Sybil-resistance principle.

The way PoW resolves forks is the rule of choosing as canonical the chain with the most accumulated work (commonly called the "longest chain"). Even if two miners create blocks almost simultaneously and a temporary fork appears, once the next block attaches to one side, that chain's cumulative difficulty grows and everyone converges onto it. For this reason confirmation is probabilistic. The deeper a block is buried (the more blocks pile up behind it), the exponentially smaller the chance it is reversed, and in Bitcoin practice 6 confirmations (about one hour) is treated as de facto finality.

Even as the network's total hash power rises or falls, Bitcoin automatically adjusts difficulty every 2,016 blocks (about two weeks) to keep the block interval at about 10 minutes. Mining rewards also halve every 210,000 blocks (about four years), so that after the 2024 halving 3.125 BTC is paid per block, causing the money supply to converge. This design intricately weaves together "economic incentive (reward) and physical cost (electricity)," and the fact that it has protected value worth thousands of trillions of won for over 15 years without a single catastrophic consensus failure attests to PoW's robustness. Yet in return, the Bitcoin network consumes on the order of about 100 TWh per year (comparable to the electricity consumption of a small or mid-sized country), carrying a sustainability controversy.

Meanwhile, because PoW is designed as "one vote = compute," it produced the paradox of concentration into mining pools. As individual miners found it hard to endure the variance of being rewarded once every several years, they gathered into pools that share revenue, so that a few large pools came to hold a substantial share of the network's hash power. This is a phenomenon of being "physically distributed yet concentrated in decision-making," teaching that decentralization must be viewed not as a number but from the perspective of effective control. The 2014 incident, in which one Bitcoin pool momentarily approached 50% and the community voluntarily dispersed its hash power, is a representative case showing the vulnerable point of PoW decentralization.

B. Proof of Stake (PoS)

Proof of Stake (PoS) binds voting power not to computing power but to the stake deposited (staked) as collateral in the network's tokens. A block proposer (validator) is selected with probability proportional to its stake share, and if it violates the rules, part or all of its deposited stake is slashed. For an attacker to seize the chain, it must buy up a majority of the total stake; but if the attack succeeds, the value of those tokens plummets and slashing burns the attacker's own capital, so an economic deterrent of "the attack cost is the attacker's own assets" operates. If PoW draws cost from the outside (electricity), PoS internalizes the cost from within (tokens)—this is the fundamental difference.

PoS's greatest benefit is energy. On September 15, 2022, Ethereum switched from PoW to PoS through The Merge, reducing the network's energy consumption by about 99.95% (from roughly 78 TWh/year to around 0.01 TWh/year). To become a validator in Ethereum PoS, one must deposit 32 ETH, and time flows in units of a 12-second slot and 32 slots = 1 epoch (about 6.4 minutes). Finality is achieved by Gasper, which combines the Casper FFG finality gadget with the LMD-GHOST fork choice, and typically a block becomes "finalized" after 2 epochs (about 12.8 minutes), gaining an irreversible economic finality.

The representative hard problems early PoS research faced were Nothing-at-Stake and the Long-Range Attack. The former is the problem that, unlike PoW, signing multiple forks simultaneously incurs no extra cost, so validators support every fork and consensus fails to converge; the latter is an attack that secretly builds a longer fake chain from some past point and rewrites history. Modern PoS practically curbs these with slashing that detects contradictory signatures and burns stake, with finality that drives in irreversible checkpoints, and with weak subjectivity checkpoints that trust everything before a certain point.

The economic structure of PoS also differs from PoW in that it ties consensus and monetary policy into one. PoW compensates miners' real costs through new issuance (block rewards), whereas PoS incentivizes validators through staking rewards (interest) on deposited stake. For this reason, in a PoS network "how much to deposit to guarantee what yield" becomes the security budget of consensus itself, requiring a delicate balance: if rewards are too low, participation shrinks and security weakens; if too high, inflation dilutes token value. In practice, Ethereum is designed so that individual yield falls as total deposits rise, automatically mitigating excessive stake concentration.

C. Delegated Proof of Stake (DPoS) and Other Families

Delegated Proof of Stake (DPoS) has all token holders, instead of producing blocks directly, vote to elect a small set of representative Block Producers to whom alone block production is entrusted. EOS, which fixes delegates at 21, and Tron, at around 27, are representative cases. Because the validator count is small and blocks are produced in turn, throughput of thousands of transactions per second and sub-second finality are possible, so it is called a "representative-democracy" high-performance consensus. However, power concentrates in a few delegates, and the risk of collusion and cartels grows, exhibiting the classic trade-off within the trilemma of gaining scalability at the cost of giving up much decentralization.

Beyond these, Proof of Authority (PoA), in which a small set of trustworthy validators signs blocks on the collateral of their identity, suits enterprise consortia and testnets, and hybrids that graft the BFT family onto public networks, like Tendermint/Cosmos providing instant finality every block, are also widely used. These families are chosen according to the trust level of public versus private blockchains ([[blockchain-types]]), and should be understood not as a single "correct answer" but as a use-case spectrum.

D. Two Types of Finality

The most practical yardstick for evaluating a consensus algorithm is "when may we trust that a transaction cannot be reversed"—that is, finality. Finality divides broadly into Probabilistic Finality and Absolute/Economic Finality, and this distinction governs exchange deposit policies, payment approval timing, and rollback-contingency design.

Probabilistic finality is the characteristic of PoW/Nakamoto consensus. Even after a block is buried in the chain, it can in theory be reversed if a longer chain appears, but as more blocks pile up behind it the probability converges exponentially to 0. Hence there is no mathematical point of "absolute finality," and an operational compromise exists of setting different confirmation counts according to risk tolerance, such as Bitcoin's 6 blocks or dozens of blocks for high-value deposits at some exchanges. This is flexible but, in that "whether it is final is a probability," imposes latency and uncertainty burdens on financial settlement.

By contrast, BFT-family and modern PoS equipped with a finality gadget provide absolute finality that, the moment a quorum (two-thirds or more) of signatures gathers, nails down "everything before this block can never be reversed under any circumstances." Ethereum's 2-epoch finality is an example; to reverse it, more than one-third of the total stake would have to accept slashing, making it economically all but impossible. In short, PoW is a model that "grows safer as time passes," while BFT-type PoS is a model where "safety is fixed at a particular point," and one must choose according to the service's settlement character (small-amount immediacy vs. high-amount certainty).

3. The Block-Finalization Process of PoW and PoS

The sequence below contrasts how PoW's competitive mining and PoS's election-and-attestation-based finalization reach the same goal (agreement on a global order) by different paths.

sequenceDiagram
    participant U as "User (creates tx)"
    participant P as "Mempool (pending txs)"
    participant W as "PoW miner"
    participant S as "PoS validators/committee"
    participant C as "Ledger (finalized chain)"
    U->>P: broadcast signed tx
    Note over W: PoW path
    P->>W: collect txs into a candidate block
    W->>W: search nonce (hash < target)
    W->>C: broadcast first winning block (longest-chain rule)
    Note over C: probabilistic finality after 6 blocks
    Note over S: PoS path
    P->>S: proposer elected by stake proportion
    S->>S: committee attests (votes) on the block
    S->>C: finalized by cumulative votes over 2 epochs
    Note over C: economic (absolute) finality

In PoW the core driving force is the competition of "whoever solves first gets to write," while in PoS it is the election + voting of "the chosen one writes and the majority attests." As a result, PoW's confirmation is an asymptotic guarantee whose reversal probability shrinks as time passes, whereas PoS's finality is a discrete guarantee that at a specific point nails down "up to here can never be reversed"—a qualitative difference. This difference is practically crucial in exchanges' deposit-confirmation policies (PoW coins require more confirmation blocks) and in assessing rollback possibility.

In both paths the block proposer holds the power to decide in what order to include mempool transactions, and here arises a common problem of MEV (Maximal Extractable Value), extracting profit by manipulating transaction order. For example, front-running by inserting one's own transaction ahead of a large buy order is typical. Because this stems, regardless of the kind of consensus algorithm, from "who holds the ordering power," complementary designs that check that power by separating block construction from proposal, like proposer-builder separation (PBS), are discussed together on both the PoW and PoS sides. If consensus's safety keeps the ledger "unalterable," MEV mitigation is another axis of trust that makes the "ordering fair."

4. Comparison of Consensus Algorithms and Selection Criteria

The table below is a supplementary summary; more important than the numbers is why those differences arise.

Category PoW PoS DPoS BFT family (PBFT, Tendermint)
Sybil-resistance resource compute (electricity) staked stake delegated stake pre-approved identity
Finality probabilistic (e.g., 6 blocks) economic (~12.8 min) sub-second instant (per block)
Throughput low (~7 TPS) medium high (thousands TPS) high (small set)
Decentralization high high–medium low low (permissioned)
Energy very high very low very low low
Representative cases Bitcoin Ethereum EOS, Tron Cosmos, Hyperledger

The numerical differences in the table stem from design philosophy. PoW's low throughput (Bitcoin about 7 TPS) is the result of a decentralization-first choice—"if nodes worldwide are to verify and sync every block, blocks cannot be made frequent or large"—not a defect. Conversely, DPoS achieves thousands of TPS because it reduced the validating parties to dozens, lowering communication cost, at the price of greater collusion risk. Thus the essential question of selection is not "which is superior" but "which of decentralization, finality speed, and throughput can this service afford to give up."

As a concrete case, the cost of a 51% attack makes the difference stark. In PoW, one must lease or hold a majority of the total hash power, so at Bitcoin's scale the facilities and electricity alone run into the trillions of won and are unrealistic; but for small PoW coins with little hash power (e.g., the former Ethereum Classic), attacks via hash leasing actually succeeded several times, causing double-spending damage on the order of billions of won. In PoS, one must buy up a majority of the total stake, during which the price soars, and on a successful attack slashing burns the capital, so an attack is tantamount to suicide. The reason financial permissioned chains choose the BFT family is also clear—among a few mutually identified institutions there is no anonymous Sybil threat, so quorum voting that gives instant finality is reasonable instead of costly PoW.

Another selection axis is governance and ease of change. PoW chains, with the interests of miners, nodes, and developers dispersed, change their protocol slowly and conservatively (Bitcoin's sluggish feature additions are an example), giving a strength in "unchanging store of value." PoS/DPoS can push upgrades quickly through stake-based on-chain governance, suiting an "evolving platform," but carry the risk that a few large stakeholders dominate decisions. Therefore consensus selection must be a comprehensive judgment that considers, beyond performance and energy, the change speed and power distribution that the chain aims for.

5. Deep Dive — Latest Trends and Likely Exam Directions

Since the PoS transition, the core task of Ethereum's roadmap is shortening finality latency. Research on Single Slot Finality (SSF), which aims to ultimately finish within one slot (12 seconds) the finality that currently takes about 12.8 minutes, is underway, and in 2025 3-Slot Finality (3SF), which splits voting across three slots, was proposed as a realistic interim step. Also, the Pectra upgrade of May 2025 raised the validator's effective-balance cap from 32 ETH to 2,048 ETH (consolidating validator counts), laying a stepping stone toward SSF. That said, SSF/3SF are still at the research/proposal stage, so it is premature to assert a mainnet adoption date.

The Modular Blockchain trend that separates the consensus layer from the execution layer is also noteworthy. In the past a single chain bore consensus, execution, and data availability all at once, but recently a structure has taken hold in which consensus and settlement are entrusted to a base layer like Ethereum while execution is handled in parallel by hundreds of rollups. Here the base layer's PoS consensus plays only the role of an "anchor that finalizes the ultimate truth," and throughput is scaled at the upper layers, thereby bypassing the scalability bottleneck of the trilemma through layer separation. That is, the center of gravity is shifting from a performance race of the consensus algorithm itself to an architecture race of "where to place consensus."

Changes on the regulatory and sustainability side are also conspicuous. PoW's enormous power consumption led to environmental-regulation pressure in the EU and other regions, which became an external driver accelerating the PoS transition. At the same time, PoS's stake concentration (validators clustering into large staking services and exchanges) and the systemic risk of liquid staking tokens (LSTs) have emerged as a new decentralization controversy. In short, the key of the latest flow is that consensus's challenge is shifting from "energy" to "stake/validator concentration." In an exam answer, showing this dynamic perspective—that the trilemma's bottleneck shifts over time—is the high-scoring point.

Likely exam directions include: ① compare PoW and PoS on Sybil resistance, finality, and energy, and discuss selection criteria; ② explain the principle of the 51% attack and the difference in defense under PoW and PoS; ③ describe the trade-offs of consensus algorithms from the perspective of the blockchain trilemma. Answering by connecting how public-blockchain consensus differs from classical BFT ([[byzantine-fault-tolerance]]) and distributed consensus ([[distributed-consensus]]) reveals depth.

6. Considerations and Implications (Professional Engineer's Perspective)

  • Application strategy — start from the trust model. For an enterprise consortium whose participants are mutually identified, it is reasonable to obtain instant finality and high performance with the PoA/BFT family instead of costly PoW/PoS. Conversely, for a public/financial public ledger where anyone must participate without censorship, PoS that preserves decentralization becomes the default. Consensus selection must start not from technology but from the governance decision of "whom do we trust and how much."

  • Trade-offs — the trilemma is unavoidable. Raising throughput reduces validators and sacrifices decentralization; preserving decentralization slows finality. Rather than trying to maximize all three axes at once, the designer must explicitly define the concession point the service can tolerate and pursue in parallel strategies that bypass the bottleneck through layer separation such as rollups and sharding.

  • Outlook — finality latency and stake concentration are the next battlefield. Now that the energy problem has largely been resolved by PoS, the focus of technical competition is shifting to "how fast can we finalize irreversibly (SSF)" and "how to distribute validation power so it does not concentrate in a few." Applying quantum-resistant signatures across public networks (e.g., [[blockchain-types]]) to prepare for the quantum-computing threat is also a mid-to-long-term task.

  • Review/audit perspective — inspect consensus parameters. From the standpoint of a professional engineer or auditor, when evaluating a blockchain-adoption project one must not stop at "which consensus is used" but concretely verify the adequacy of consensus parameters such as validator count, stake distribution, slashing conditions, finality latency, and upgrade governance. In particular, one must beware of dressing up DPoS/PoA that rely on a few validators as "decentralized," and the key is to require documented proof of the alignment between service requirements (finality speed, censorship resistance, regulatory compliance) and the actual consensus characteristics.

  • Related technologies and risk management. Consensus does not stand alone. One must design and audit the end-to-end trust chain as a whole—the oracle that brings in external data ([[blockchain-oracle]]), the layer-2 rollups for scaling, the financial risk of staking derivatives, and even exchange deposit-confirmation policies (contingency for rollback of probabilistic-finality coins). In particular, the precedents of 51% attacks on small PoW chains left the lesson that "consensus safety is proportional to network size (total hash/stake)."

References


In one line: Blockchain consensus is a mechanism that, in a trustless anonymous environment, binds voting power to scarce resources such as compute (PoW), stake (PoS), and delegation (DPoS) to deter Sybil attacks and converge on the ledger's global order, and it must be chosen to fit the trust model amid the trilemma of decentralization, security, and scalability.