Decentralized storage has been a concept for many years, but few truly understand the underlying technical differences. Recently, I tested the mainstream solutions on the market one by one—ranging from Filecoin's miner mechanism to Arweave's permanent storage design, and then to IPFS's routing architecture—only to find that Walrus's RedStuff encoding scheme truly finds the balance between efficiency and cost.

The Filecoin project caused quite a stir during its 2017 ICO, and I also bought some FIL on the hype. Only when I used it did I realize how idealistic it is. The separation architecture of storage miners and retrieval miners sounds clever, but in practice, it split into two markets. Storing a file requires negotiating with storage miners, while retrieving it involves finding retrieval miners, with various proof mechanisms in between. Especially for small files, it's a nightmare—miners prefer large orders, and the cost for small files becomes ridiculously high. I once tried storing a 500MB dataset, and it took half a day just to find miners; in the end, I gave up.

Arweave's concept of permanent storage is indeed attractive—pay once, store forever. But if you do the math with a calculator, you'll see the costs clearly. It uses a full network replication model, meaning each file must be fully stored by every node in the network. The larger the network, the greater the storage burden on each node. To spread out these long-term costs, the pricing is directly set at $9,700 per TB. Archives or important documents might be acceptable, but for large-scale data like AI training datasets, the costs completely spiral out of control.