Home

Hidden Geographical Fractal Ring (HGFR)

Interface/API provided for the Adapter:

• broadcast(Message m)
• join(NodeProfile np)

Interface/API provided for the Wire Protocol Layer

• send(Packet p, String sender_ip, String receiver_ip)

---

Observation on P2P Network:

1. Reach all nodes (sufficient amount of nodes); [fault-tolerance/robustness]
2. Secure/address threat models; [security]
   • transport correct message (immune to falsifying msg, addressed by SGX);
   • hide the topology (anonymity);
   • malicious individuals (don't consider message changing on the wire), eclipse attack, Sybil attack, DDoS attack;
3. Converge as fast as possible (route should be optimal & msg should have no redundancy); [efficiency]

---

Proof & Analysis of HGFR

Hidden Pattern/Topology

• [packet] send/out (behavior the same)
• [packat] receive/in (contact node receive info from upper-layer ring and lower-layer ring, so send fake message to prev-contact nodes in upper/lower-layer rings)
• [time] change contact node periodically, not long enough for the attacker to statistically infer the difference

Reach sufficient # of nodes

• hidden topology -> no specific target found -> random failure makes sense
• fault ratio analysis
  • each ring contains r nodes, and each ring elects c contact nodes;
  • random failure:
    • ours: probabilitic reaching all remaining nodes with a sufficient probability;
    • gossip: storge and communication overhead to reach all remaining nodes (with well connected graph);

Efficiency Analysis

• message complexity
  • Gossip: O(Nm), where m is the routing table size;
  • HGFR: O(N);
• time complexity (communication rounds)
  • Gossip: O(N/r*logr), where r is the number of nodes on the ring;
  • HGFR: O(logN);

---

Evaluation

Settings:

• Traffic Control, to simulate real RTT, sleep after receiving packets if between:
  • continents: 150~200ms (random)
  • countries: 100~150ms (random)
  • cities: 50~100ms (random)
  • district: 20~50ms (random)
  • within the same district: do not sleep;
• ID -> id|district|city|country|continents
• Solve docker problem: multi-swarm;
• Blockchain workload generator (oringinal version);

Comparison Targets:

• Kademlia (how to compare? simulate encryption-decryption time)
• Chord
• Pastry
• Tapestry

Time Complexity

Fixed node number: Time / # of messages/operations

Message Complexity

Fixed node number: # of messages sent / # of operations

Scalability

• Time / # of nodes (fixed # of messages/operations)
• Message / # of nodes (fixed # of operations)

Fault-tolerance

• Message Reachibility / Fault Ratio
• Time (extra message complexity) / Fault Ratio

Anonymity

• Wireshark packet watching in an interval, compare particular node & normal node

Maintainability

• Node Loss Recovery Time / # of existing nodes
• Node Join Boostrap Time / # of existing nodes