Content-Addressable Storage

Content-addressable storage (CAS) is a data storage paradigm in which an object's address is derived from the object's content, typically by applying a cryptographic hash function such as SHA-1 or SHA-256. The resulting digest is used both to store the object and to look it up later. Because the address is a function of the bytes themselves, two clients that hold the same content arrive at the same identifier without coordination. Three properties fall out of this design. First, deduplication is automatic: writing identical content twice maps to the same address, so the underlying store keeps a single copy. Second, objects are effectively immutable — modifying the bytes produces a new address, leaving the original unchanged and addressable. Third, integrity verification is intrinsic: a reader can rehash the retrieved bytes and compare against the address to detect corruption or tampering. Together these make CAS well suited to fixed content, archival data, and distributed systems where parties do not trust each other's storage layer. The term emerged in the late 1990s. Paul Carpentier and Jan van Riel of the Belgian startup FilePool coined it while building a platform for compliance archiving; EMC acquired FilePool in 2001 and shipped the technology as Centera in 2002, marketed as a write-once, read-many store for records that regulations such as Sarbanes-Oxley required to remain unaltered. Around the same time, Sean Quinlan and Sean Dorward of Bell Labs published Venti, an archival block store for Plan 9 that addressed every block by its 160-bit SHA-1 digest and enforced a write-once policy at the protocol level. Content addressing then spread beyond enterprise archiving. Git (version control), released by Linus Torvalds in 2005, models every blob, tree, commit, and tag as a content-addressed object; the project is gradually transitioning from SHA-1 to SHA-256 to harden it against collision attacks. IPFS, introduced by Juan Benet in 2014, generalizes the idea into a peer-to-peer file system whose objects are nodes in a Merkle DAG identified by self-describing multihash content identifiers (CIDs). Other examples include the CAStor object store and the content-addressable layers inside container registries such as Docker and OCI. The same mechanic — hash-as-name plus an immutable graph of references — also underlies Merkle-tree constructions in blockchains and certificate transparency logs, making CAS one of the foundational ideas of modern distributed data infrastructure.