How does it work?

Let's look at an overview of how IPFS works. When we store a file in IPFS, it's split into chunks < 256 KB and hashes of each of these chunks are generated. Nodes in the network hold the IPFS files they need and their hashes in a hash table.

There are four types of IPFS files: blob, list, tree, and commit. A blob represents a chunk of an actual file that's stored in IPFS. A list represents a complete file as it holds the list of blobs and other lists. As lists can hold other lists, it helps in data compression over the network. A tree represents a directory as it holds a list of blobs, lists, other trees, and commits. And a commit file represents a snapshot in the version history of any other file. As lists, trees, and commits have links to other IPFS files, they form a Merkle DAG.

So when we want to download a file from the network, we just need the hash of the IPFS list file. Or if we want to download a directory, then we just need the hash of the IPFS tree file.

As every file is identified by a hash, the names are not easy to remember. If we update a file, then we need to share a new hash with everyone that wants to download that file. To tackle this issue, IPFS uses the IPNS feature, which allows IPFS files to be pointed using self-certified names or human-friendly names.