I've been working on the comparative testing of data availability layers these past few days, and I conveniently migrated a few large JSON files that were previously running on Arweave to the Walrus testnet. To be honest, the narrative of 'permanent storage' emphasized by Arweave often turns out to be a false proposition when it comes to practical engineering, especially for dynamic NFT metadata that requires frequent updates; the one-time buyout cost model has become a burden instead. After running the Walrus CLI tool, the biggest difference I felt is that its handling of Blob data is more like Web2's S3, rather than that kind of monster that sacrifices efficiency for decentralization.
Last night, I stayed up all night testing the recovery ability of Erasure Coding when nodes drop out. I deliberately disconnected two storage nodes in the local environment, and the delay in data reconstruction was almost negligible. This is much lighter than Filecoin's complex sector packaging and proof mechanism, and Filecoin's retrieval market has yet to run smoothly; retrieving data is as slow as dial-up internet. The approach of Walrus, which decouples storage and computation, clearly aligns better with the current expansion pace of the Move ecosystem.
However, there are some parts of the testnet documentation that are quite obscure, and the parameter configurations didn't align, so I had to go to Discord to review the chat records to get it running. At this stage, it doesn't seem like it aims to overthrow Arweave but rather fills the vacuum between high-performance, temporary, and low-cost storage. This pragmatic engineering orientation is, in fact, more grounded than those projects that tout 'the eternal preservation of human knowledge.'
