I plan to write documentation of the IPFS process including the PXE router config later at https://github.com/majbacka-labs/nixos.fi -- we might also run a small public build server for peoples Flake configs, who are interested in trying out this process.
You're doing some cool things here.
Remember in the late 90's booting server off a CD-ROM was the thing.
But you shouldn't need to: you should be able to do the same thing with a docker graph driver, so there is no registry - even daemon should perceive the local registry as "already available", even though in reality it's going to just download the parts it needs as it overlay mounts the image layers.
Which would actually potentially save a ton of bandwidth, since the stuff in an image is usually quite different to the stuff any given application needs (i.e. I usually base off Ubuntu, but if I'm only throwing a Go binary in there plus wanting debugging tools maybe available, then in most executions the actual image pulled to the local disk would be very small).
"Kraken was initially built with a BitTorrent driver, however, we ended up implementing our P2P driver based on BitTorrent protocol to allow for tighter integration with storage solutions and more control over performance optimizations.
Kraken's problem space is slightly different than what BitTorrent was designed for. Kraken's goal is to reduce global max download time and communication overhead in a stable environment, while BitTorrent was designed for an unpredictable and adversarial environment, so it needs to preserve more copies of scarce data and defend against malicious or bad behaving peers.
Despite the differences, we re-examine Kraken's protocol from time to time, and if it's feasible, we hope to make it compatible with BitTorrent again."
I created Spegel to fill the gap but focus on the P2P registry component without the overhead of running a stateful application. https://github.com/spegel-org/spegel
Fully distributed OS's/Virtual Machines/LLM's/Neural Networks
If LLM's are token predictors for language, what happens when you do token prediction for computation across a distributed network? Then run a NN on the cache and clustering itself? Lots of potential use cases.