How IPFS Works: A Deep Dive into Its Architecture and Core Components

Overview

IPFS is tightly linked to blockchain and addresses the growing need for off‑chain data storage by using content‑addressed, encrypted blocks stored outside the chain. The article examines IPFS’s design from the source code, noting that the implementation evolves over time.

What Is IPFS?

IPFS is a distributed file system protocol that enables persistent storage and sharing of files. It differs from traditional file systems by using distributed storage, splitting files into blocks identified by unique hash IDs, and employing content‑addressing to retrieve blocks.

Blocks are organized into a Merkle‑DAG, providing content addressing, tamper resistance, and deduplication.

IPFS System Architecture

The architecture consists of five layers:

Naming – a PKI‑based namespace.

MerkleDAG – the internal logical data structure.

Exchange – the protocol for block data exchange between nodes.

Routing – implements node and object addressing using the KAD algorithm.

Network – encapsulates P2P communication and transport.

From a data perspective, IPFS comprises two major modules:

IPLD (InterPlanetary Linked Data) for data definition and modeling.

libp2p for data transport.

IPLD

IPLD provides a unified data model that enables content addressing across domains (e.g., blockchain, Git). It defines concepts such as merkle links, merkle‑DAG, merkle‑path, CID, data model, serialization format, and selectors. IPLD uses multiformats (multihash, multiaddr, multibase, multicodec, multistream) to ensure extensibility and upgradeability.

CID (Content Identifier) is the core identifier in IPFS, with two versions. CIDv0 uses base58btc, protobuf‑mdag, and multihash. CIDv1 adds multibase, version, multicodec, and multihash components for greater flexibility.

libp2p

libp2p is a modular network protocol stack that separates routing, swarm (transport and connection), distributed record store, and discovery.

Routing implements KAD and MDNS routing; KAD uses XOR distance to organize nodes into K‑buckets.

Swarm defines transport, connection, and stream multiplexing interfaces, with dynamic stream protocol negotiation via multistream‑select and ls messages.

Distributed Record Store stores key‑value records, enabling IPNS name publishing and resolution.

Discovery supports bootstrap, random walk, and MDNS methods to find peers.

Conclusion

The article summarizes that IPLD defines and models data while libp2p handles its transport; both can be used independently. IPFS aims to replace HTTP, and integrates Filecoin incentives to improve data persistence.