Bayesian Persuasion

GNUnet notes

3 min read

The Architecture of the GNUnet: 45 Subsystems in 45 Minutes (2015)

He gives 45 things you can do with GNUnet (in 2015) that try to emphasize current capabilities. He does not go into how they work

  • all processes managed around a systemd-like process called ARM (automated Restart Manager)
  • lowest level is transport. Below we don’t care (TCP, UDP, Bluetooth, whatever)
    • unreliable, out-of-order packet delivery semantics
    • enforces connection restrictions
    • has an ATS (automated Transport Selection) to decide which connections to establish
      • allocates bandwidth to peers by network topology
      • allows other subsystems to specify preferences
  • CORE: off-the-record link encryption between peers
  • HOSTLIST: allows for bootstrapping → multiple methods for this
  • Network Size Estimate (NSE): byzantine fault tolerant (HOW??)
    • all peers converge to same network size estimate
  • DHT:
    • Tolerates small-world underlay topology
    • optionally track path key-values took in the network
    • key difference is that it does make the assumption that every peer can talk to every other peer
    • plugins to verify integrity of key-value pairs as there can be multiple values for a given key
      • e.g. this key should be hash of value
  • Confidential Ad-Hoc Decentralized E2E Transport
    • AXOLOTL-encrypted E2E comms
    • reliable or unreliable
    • in or out of order
    • low-lat or buffered
  • Identity (management)
    • public key pairs as “egos” to identify users
    • each user can have multiple alter-egos
    • separate from peer identities (network addresses)
  • GNU Name System (GNU): secure memorable names
    • interoperable with DNS
    • key: achieves query and response privacy (HOW???)
  • Zone management: labels and record types
  • Key revocation protocol:
    • revocation messages can be prepped and stored off-line → highly efficient
  • Set: compute set union or set intersection
    • surprisingly low bandwidth required
  • Scalarproduct service (SMC): secure multiparty computation
    • if one party lies, result is garbage
  • Random Peer Sampling: BFT, decentralized, (Braums???)
  • Multicast (WiP):
    • source not required to key-exchange with each group member (must trust members of each group)
  • PSYC2: runs over Multicast → extensible messaging format
  • Social (Network Applications): runs over PSYC2 with GNS
    • key concepts: nym, place, host, guest
    • SecuShare (example of this)
  • Statistics: collect numeric run-time info
  • TestBed infrastructure to run controlled experiments to collect Statistics
  • Conversation: User application for phone calls (E2E encrypted)
  • File-”sharing”: prefer to call it “publishing”, lol
    • Merkle tree of blocksk from a file
    • peers caching cannot view contents
    • multi-source download
    • keyword search (???)
    • file meta-data available as part of search result
    • can share directories and mount via FUSE
  • Search by REgular EXpression
    • fully decentralized, uses DHT
    • trivial api, but very non-trivial theory
  • can use this to do DNS translation
    • or IP-over-GNUnet
  • BFT Consensus
    • synchronous
    • compute global union over a set of initial elements distributed across n-k honest participants
    • all honest end up with exact same set
    • final set is super-set of union of initial elements at honest peers
  • using this, get electronic voting
    • implemented Cramer’97-style e-voting
      • correctness, secrecy, indi verif, univ veri, fairness, robustness, coercion res.
      • participants: supervisor, authorities, voter
  • RESTful APIs (WiP):

Graph View