4 ms·
I absolutely love the concept and the underlying moral reasoning for why we need IPFS! One of the first questions that came to me: Is there any way to add a fo
by PhaseMage 11y ago
I absolutely love the concept and the underlying moral reasoning for why we need IPFS!
One of the first questions that came to me: Is there any way to add a forward error correcting code, like an erasure code to IPFS? I didn't find any discussion of this in the IPFS paper.
This seems somewhat critical to be able to compete with modern centralized storage systems which are likely to use FEC extensively to provide the level of redundancy that customers expect. Modern FEC codes can provide phenomenal levels of redundancy with less than 50% overhead. IPFS seems to rely on many fully-redundant copies?
- crypt0x 11y agoInteresting Idea! I think TAHOE-LAFS also uses something like this? From a practical standpoint, I'd like my local node to have a full copy so that my friends coming over don't need to reach out to the (maybe offline) internet but I don't see why we shouldn't support encoding the chunked blocks in such a way?
- dgoldstein0 11y agoBased on my reading of this article, IPFS doesn't specify how you store the content - just that the content you provide back matches the requested hash. So in theory this could easily be implemented under the hood. Though perhaps I'm misunderstanding your question - if you want multiple IPFS nodes to coordinated on storing parts of an erasure encoded file? that would be useful too. Again I think it's possible - you could build a small distributed system to host files this way. The most interesting variant though is if IPFS could have support for something like this natively - so nodes run by different users can each pitch in a little. Conceptually it's possible, but probably the protocol doesn't support it.
- fla 11y agoReminds me how Usenet binaries are often posted along with par2 files. Reed-Solomon ECC are just amazing!