Asked because I saw no mention of erasure coding. IPFS never worried about availability.
Afaik stronger availability always comes from the interaction between erasure coding and distribution, so erasure coding cannot live in too separate a layer from distribution choices.
At the same time, there should be good strong-ish availability regimes based upon fountain codes instead of reed-solomn, which wind up much weaker than the expensive strict 2/3rd honest one used by polkadot, or the aggressive sampling used by celestia, but still much stronger than simple replication like ipfs protocols, bittorrent, etc. Yet afaik nobody formulated one. It does matter to have the right storage abstraction as doing so may help to formulate this.
I’m curious how Iroh and/or ipfs compare with libtorrent (uTP) at raw transfer speed under different network conditions. I’d expect libtorrent blows them away under most conditions, but not under all conditions due to LEDBAT.