[stavros]

IPFS is a good solution in theory, but in my experience it works so poorly in practice that I wouldn't trust it as my only form of archival. Certainly add files to IPFS too if you want, but not exclusively.

[kkielhofner]

I second this and for anything other than "need to fetch a file or two occasionally" it's essentially broken in my mind. Expect to wait anywhere from 3 - 30 seconds for first byte. When modern CDNs are doing first byte in 100ms (or whatever) over HTTPS that's an eternity for most users.

Virtually all IPFS storage and bandwidth from pinning and gateway providers is at least 2-3x what you'd expect to pay for S3, etc (because they just use S3 on the backend). The public IPFS gateways have such low request limits and especially poor performance they often struggle to load a website with more than a couple of IPFS hosted assets on it.

If you want to run your own node go-ipfs is extremely difficult if not impossible to use at anything more than toy scale. Eats RAM like crazy, uses a TON of bandwidth (not unexpected but still seems like a lot), garbage collection is broken, and much more...

Frankly the state of IPFS is embarrassing considering it's seven year old tech.

[dannyobrien]

For the delay do you mean via the web gateways, or directly over IPFS?

For storage, most of the Filecoin-backed serves (like http://web3.storage, https://nft.storage, and https://estuary.tech are either free or close to that for most use cases. Have you looked at those recently?

I hear you on go-ipfs, but there are now multiple implementations, include rust-ipfs, that are also getting there.

[akavel]

Is Dat/hypercore or other similar alternative more reliable in practice? Are torrents better? (But wouldn't tracker hosting be a "chicken vs. egg" issue?)

[stavros]

Hmm, I remember Dat being a bit more reliable when I tried it, but I have used IPFS way more. Torrents are good too, but you can't update them, which might be an issue.