IPFS operates more like a distributed version of HTTP, where the concept of "writing" to it isn't quite the same as traditional file systems. When you add or modify content in IPFS, it generates a new Content Identifier (CID). This unique CID is deterministic, meaning if two users have the exact same data, they will generate the same CID. This characteristic is what enables the deduplication feature in IPFS. For instance, if a single node has two directories containing identical copies of a file, IPFS recognizes it as one file with a single CID, thus not duplicating the storage.

To manage updates and modifications efficiently, IPFS incorporates additional layers like IPNS (InterPlanetary Name System), ENS (Ethereum Name Service), orbitdb, or the Mutable File System (MFS). For example, you can use the MFS to append to a file by using the command ipfs files write --help for append mode. In practice, you might link a file into your MFS with a command like ipfs files cp /ipfs/Qmfoo /a.

It's important to understand that you don't "write to IPFS" in the traditional sense. Instead, IPFS is a protocol for serving content, where each piece of content is accessed via its unique CID. This approach ensures efficient content retrieval and storage, leveraging the benefits of deduplication and content-addressability. For a quick and concise explanation of how IPFS works, you can watch this 4-minute breakdown: IPFS in 4 minutes.