cloudparts

Distributed ledgers

A record that several organisations keep in step without any one of them owning it: each runs its own copy, and changes are signed, agreed through consensus and chained so the history can’t be quietly rewritten.

2 technologies on this page ↓

How it works

A distributed ledger is a replicated, append-only log of transactions plus the state that results from applying them, run by nodes that belong to different owners. Clients sign each transaction with a private key. Nodes agree on one order for transactions through a consensus protocol, group them into blocks, and link each block to the previous one by including its hash, so changing an old entry would change every hash after it and every other copy would notice. Each node applies the same transactions in the same order, often through small programs called smart contracts, and so reaches the same state. The big split is who may take part. On a permissionless ledger such as Bitcoin or Ethereum anyone can run a node and propose blocks, so the protocol has to make lying expensive: proof of work burns electricity, proof of stake puts deposited money at risk, and a native cryptocurrency pays for both. On a permissioned ledger such as Hyperledger Fabric, or Besu running a private network, every node and user has an identity issued by a member organisation, and a known set of validators agrees on order by voting, with no token: Byzantine fault-tolerant protocols such as QBFT or SmartBFT keep working while fewer than a third of them misbehave and give final answers within seconds, and Fabric can also use crash-tolerant Raft when the ordering nodes are trusted not to lie. Either way the cost of trust shows up as speed: every write is replicated to, checked and stored by many parties that don’t trust each other, so a ledger handles far fewer transactions than a single database with the same hardware, and data written to it is hard to correct or delete.

Think of it as

A shared paper ledger in a village where every family keeps an identical copy. A new entry only counts once enough families have checked the signatures and written it into their own books in the same order, and each page starts by copying a fingerprint of the page before. Nobody needs to trust the clerk, because there is no clerk, but every purchase takes a trip round the village.

Reach for it when

  • Several organisations must share one record, none of them will accept another (or a vendor) as the operator of a central database, and a neutral third party isn’t available or affordable
  • Every party must be able to check the full history for itself, for example to prove to a regulator or a court that nobody altered a record after the fact
  • Assets or obligations move between organisations and settlement currently needs reconciling separate books, such as tokenised securities or trade documents passed along a chain of companies
  • Working with public blockchains: running your own Ethereum node, or a private network that uses the same contracts and tools

Think twice when

  • One organisation writes the data, or all the writers already trust one operator: a database with audit logs, or one with built-in tamper evidence such as SQL Server ledger tables, is faster, cheaper and easier
  • You need high write rates, low latency or the ability to correct and delete records, including personal data that may have to be erased
  • The hard part is getting competitors to join, not the technology: a consortium network is worthless without its members, which is what ended TradeLens
  • The data comes from outside: a ledger proves that an entry hasn’t changed since it was written, not that it was true when someone typed it in

A real example

Five banks want a shared register of syndicated loans so they stop reconciling five copies by email. They first ask whether one of them, or an agent bank, could run a database the others trust; they can’t agree, so they set up a Hyperledger Fabric network on which each bank runs two peers and one ordering node using SmartBFT. A loan’s smart contract requires endorsements from the lending banks before a transfer is valid, and a private data collection keeps each bank’s pricing to itself, with only a hash on the shared channel. Transfers are final within seconds instead of after days of reconciliation. The cost: five teams operating certificate authorities and nodes, every contract change needing approvals from enough banks, and throughput in the hundreds of transactions per second, which a loan register can live with.

Technologies

Specific products, and what each does differently.

What each provider calls it

AWS
Amazon Managed Blockchain ran Hyperledger Fabric networks and Ethereum, Bitcoin and Polygon nodes, but takes no new customers from October 29, 2026 and shuts down on September 29, 2027; AWS points Fabric users to another provider, self-managed Fabric on EC2 or EKS, or a plain database. Its ledger database, Amazon QLDB, ended on July 31, 2025
Google Cloud
No managed permissioned ledger; Blockchain Node Engine runs fully managed Ethereum nodes on Mainnet or testnets
Azure
Azure Blockchain Service was retired on September 10, 2021, with users sent to ConsenSys’s Quorum Blockchain Service. Azure confidential ledger, an append-only ledger run in hardware enclaves on the open-source Confidential Consortium Framework, stores tamper-evident records and digests for Azure SQL ledger tables
Open source & others
Hyperledger Fabric and Besu (LF Decentralized Trust, Apache 2.0), plus Hiero and Iroha, and Cacti for linking ledgers; R3’s Corda; Microsoft’s Confidential Consortium Framework; public Ethereum clients such as Geth, Nethermind and Besu. Hosted nodes and networks from vendors such as Kaleido

Shared database or shared ledger?#

Adapted from Wüst and Gervais’s 2018 paper “Do you need a Blockchain?”. Most situations end in one of the first two rows.

SituationUseWhy
One organisation writes the dataA databaseNobody else has to agree on what was written or in what order
Several write, and all would accept one operatorA database run by that operator, with audit logs or ledger tables for tamper evidenceA trusted party that is always online does the ledger’s job faster and more cheaply
Several write, no operator they all trust, all writers knownA permissioned ledger such as Fabric or a private Besu networkIdentities replace mining, and voting among known validators gives final answers in seconds
Anyone may writeA permissionless chain such as EthereumOnly proof of stake or work, paid for with a token, keeps unknown writers honest
Outsiders must be able to verify the historyA public permissioned ledger: members write, anyone can readPublic verifiability is the one thing a members-only network can’t give
  • Even when the answer is a ledger, the members still have to agree on governance (who runs nodes, who approves contract changes, who pays) before the technology helps.

Often confused with

Coordination servicesRelational databasesNewSQL and distributed SQLBitemporal modeling
Where it came from1982–2026

The pieces are old: Leslie Lamport and colleagues described Byzantine fault tolerance in 1982, and hash-chained timestamping dates from 1991. In October 2008 the pseudonymous Satoshi Nakamoto combined them with proof of work in Bitcoin, a ledger anyone could join without asking permission, and Ethereum added general-purpose smart contracts when it launched in July 2015. In December 2015 the Linux Foundation announced Hyperledger, with banks and technology companies, to build ledgers for businesses that know each other and want shared records without a cryptocurrency.

The years from 2016 to 2019 were the boom. IBM and Digital Asset’s Fabric reached 1.0 in July 2017, ConsenSys released the Pantheon client (later Besu) for public and private Ethereum, and AWS and Microsoft launched managed blockchain services. Many projects then stalled on the problem the technology didn’t solve: getting rivals to join and agree. Maersk and IBM shut down TradeLens, a Fabric network for shipping, in early 2023, saying it had not reached commercial viability without full industry participation, and in November 2022 the Australian Securities Exchange wrote off up to AU$255 million on a ledger-based replacement for its settlement system after a review found that, with the exchange still the central source of truth, the design lost most of the benefits of a ledger.

Cloud providers have retreated. Microsoft retired Azure Blockchain Service in September 2021, AWS ended QLDB in July 2025 and announced that Amazon Managed Blockchain will close in September 2027. The open-source projects carry on under LF Decentralized Trust, formed in 2024, increasingly aimed at tokenised assets and public Ethereum rather than generic consortium databases.

Why it’s built this way

Consensus between owners, not just replicas
Ordinary replication assumes every node runs your code honestly. A ledger’s nodes belong to different organisations, so it uses Byzantine fault-tolerant consensus or proof of work or stake, which keep working when some nodes lie.
Hash-chained, append-only history
Linking each block to the last by hash makes any rewrite of history visible to every other copy, which is what lets parties rely on records they don’t control.
Permissioned networks for business
Companies that know each other don’t need mining or a token to keep strangers honest, so Hyperledger and others gave each participant a certificate-based identity and used faster voting-based consensus.
Smart contracts on every node
Running the same business rules on each copy means every party can check that a transaction followed the agreed rules, not only that it was recorded.

Timeline

  1. 1982

    Lamport, Shostak and Pease publish “The Byzantine Generals Problem”

  2. 2008

    Satoshi Nakamoto publishes the Bitcoin paper in October

  3. 2015

    Ethereum launches in July; the Linux Foundation announces Hyperledger in December

  4. 2017

    Hyperledger Fabric 1.0 is released in July

  5. 2021

    Microsoft retires Azure Blockchain Service on September 10

  6. 2022

    ASX pauses its ledger-based CHESS replacement; Maersk and IBM announce the end of TradeLens

  7. 2024

    LF Decentralized Trust takes over the Hyperledger projects

  8. 2026

    AWS stops taking new Amazon Managed Blockchain customers on October 29

Sources16

Improve this page on GitHub