network-stories/pages/Node Reputation.md

Node Reputation

Related:

• Ikingut (Reputation Algorithm)
  • Prep: [||||||||||]
  • Cook Time: X min
  • Ingredients: 1.
  • Important Conceptual Goals
    • Prep: [||||||||||]
    • Cook Time: X min
    • Ingredients: 1.
      • Simple
      • Lightweight
      • Pluggable
      • Adaptive
      • Dynamic
      • Robust
      • No Transitive Trust
      • Reasonable Bootstrap Time
  • Reputation Polling (Polling Dynamics)
    • Prep: [||||||||||]
    • Cook Time: X min
    • Ingredients: 1.
    • Indirect Request
      • Prep: [||||||||||]
      • Cook Time: X min
      • Ingredients: 1.
    • Direct Request
      • Prep: [||||||||||]
      • Cook Time: X min
      • Ingredients: 1.
      • Unirep (a pre-existing example)
      • Qualities Necessary (of Direct Request)
        • Verifiability (see Consensus)
        • Non-Repudiation (see Consensus)
        • Privacy Preserving (preserve origins of score, emit opinions without a way to trace back the origins - MPC)
    • Local Heuristic
  • Glacier Algorithm
    • Prep: [||||||||||]
      • Cook Time: X min
    • Ingredients:
      • Each Iteration
        1. Voting
        2. Agent Action (Post-Consensus Decision Finality)
      • Min-Multiplicative Reputation Punishment
        • Multiplicative
        • Linear
  • Experimental Research
    • Starting Point
      • Trust Wisdom per Node (requires further elaboration)
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Adversary Types and Effects (requires further elaboration)
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
    • Current Stage of Testing/Challenges
      • Silent-Omniscient Adversaries
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Punishment Testing (Modulating Punishment/No Punishment Impacts to Conditions)
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Adding Multiplicative-min Punishment
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Reputation Attacks
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
        • Con-Artist Attack
          • Prep: [||||||||||]
          • Cook Time: X min
          • Ingredients: 1.
        • The On-Off Attack
          • Prep: [||||||||||]
          • Cook Time: X min
          • Ingredients: 1.
      • Punishment Effects on Glacier
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Reputation Limitations
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
        • No immediate defense against coordinated attacks
        • Reputation does not add to security
    • Future Work
      • Circumstantial Impact of Reputation on Consensus
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Long-running Simulation
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Sudden changes in Collective Byzantine Behavior
        • Prep: [||||||||||]
        • Cook Time: X min
        • Ingredients: 1.
      • Pending Questions
        1. Interaction of Stake-based and Reputation-based selection
           • Stake simulations are necessary for exploring options
           • Relevant in the incentives discussion?
        2. How much (and if) does reputation really help in a coordinated attack? (Assuming patient con-artist attack)
           • Complex interactions here, this model would particularly benefit having a prototype/PoC
• Node Challenges
  • Prep: [||||||||||]
  • Cook Time: X min
  • Ingredients: 1.
  1. Intermittence Flexibility
  2. Sudden Changing Behavior
  3. Bootstrapping Quickly
  4. Resilience to Dynamic Unpredictable Network Changes (Can Handle Membership Volatility)
• Design
  • Prep: [||||||||||]
  • Cook Time: X min
  • Ingredients: 1.
  1. Local reputation view
  2. Adaptive and Dynamic
  3. Required Bootstrapping
     • Network View Not Required
     • Individual Node Reputation Not Required
• Eigentrust
  • Prep: [||||||||||]
  • Cook Time: X min
  • Ingredients: 1.
  • Transitive Trust
  • Understanding the Math
  • Trust Decay
    • Malicious Clusters (describe nuances better, trusted nodes cannot overlap the malicious collective)
  • Trust for Consensus
  • Separation of Reputation
  • Confidant
  • XRep
  • P-Grid
  • R2Trust
• Generic Taxonomy
  • Prep: [||||||||||]
  • Cook Time: X min
  • Ingredients: 1.
  • Dimensions
    • Single
    • Multiple
  • Time computation
  • Aggregation
    • Deterministic sum of positive and negative ratings
    • Probabilistic
  • Logic
    • Local vs gathered data
    • Age of data
    • Frequency of data
    • Weight of multiple dimensions
  • Value Control
    • External
    • Internal
  • Data Aging
    • None
    • Decay
    • Death of old/selected
  • Selection
    • Ranking-based
    • Threshold (trusted/untrusted)
    • Probabilistic selection