Freenet Notes

General Overview

notes on this video

General takeaways

• Seems to fundamentally be a peer connection process to maintain a healthy mesh of responding to requests
  • “more of a communication medium than a storage medium”
• should look into understanding the differentiation between IFPS data and Freenet contracts
  • it’s fundamentally just IPFS but computation instead of static data
• heavy reliance upon webassembly for computation
• it doesn’t seem to handle the adversarial environment

Raw Notes

• seems to be a VERY high level description, not going into actual mechanism

• gives description of the internet today

• Vision of freenet

• freenet core is about 5-6 MB

• fundamentally a distributed key-value store

  • key is blake3 hash of webassembly code
  • role of that code (contract)
    • validates some “state”
    • how to apply change to state
    • sync the state
  • contracts can read other contracts
  • contracts “are just data stored on the network by peers”

• allows for same API as how the web browser talks to the current internet

• how to find a given key?

  • num peers is 20-30
    • peer address is first 4 bytes of hash of IP
  • routed through network by working towards a given target between 0-1
  • greedy routing - going through the closest peer you’re connected to
  • fastest figures out fastest route, chooses it (goes over Isotonic Regression)
    • gives a implementation at https://github.com/sanity/pav.rs
  • goes over relationship between latency vs failure rate vs transfer speed
  • freenet performs regressions on its experience with the above properties to predict performance, and adjust peer connections accordingly

• bootstrapping to the network

  • initial connect request with ansatz target location, traverse, accept on optimal peer once found
  • use trick to get around firewalls (hole-punching)
  • decide optimality of peer connection via (small world network)
    • optimal: probability of connection between two peers is 1/distance
    • trying to balance connections with incoming request distribution. Does this

• resource management

  • types
    • bandwidth
    • memory
    • storage
  • consumers of resources
    • contracts
    • connected peers
  • contract’s “value” proportional to inbound requests
  • peer’s “value” is proportional to number of requests it responds to (outbound requests)
  • garbage collection of contracts with lowest value relative to cost (peer’s view)
    • similar to how IPFS serves data - performance dictated by demand

• future: data streaming

  • send response along the request chain instead of directly
    • no additional connection requirements, reuses established network topology
  • currently this must be done by sending the entire data

• questions

  • compare contrast around other similar protocols
    • freenet’s unique thing is data is identified by hash of contract
    • freenet wants to be general purpose platform for distributed computation
  • how do you get any confidence that “it works”
    • it’s difficult because an individual peer isn’t consistent (in general)
    • simulations were used to ensure topology self-organized in the way that it should (happy path)
    • then used simulated network stack to experiment and validate behavior there
    • 3rd is using something like shadow to simulate peers on a singular machine
    • then release and ensure it works as it should (very close to this as of Jun 2024)
  • contracts have state transitions, what mechanism causes eventual consistency. What pushes state transitions back towards the network?
    • when you modify the state of a contract, you start locally (mentions concept of subscription to a contract)
    • modification will be noticed by subscribers to contracts → sync protocol updates state with others
    • not a clear answer
    • contracts can (required???) define how state is updated, spread of that state transition is viral throughout the network (flooding???)
    • lots of assumptions of optimistic execution

Ghost Keys Video

• notes on this video