DHT & Peer Discovery

KayakNet uses a Kademlia-based Distributed Hash Table (DHT) for peer discovery and decentralized data storage.

What is a DHT?

A DHT is a decentralized key-value store distributed across all nodes:

• No central server

• Data is replicated

• Lookups are efficient (O(log n))

• Survives node failures

Kademlia Protocol

KayakNet uses Kademlia, the same DHT protocol used by BitTorrent.

XOR Distance

Nodes are assigned IDs, and "distance" is calculated using XOR:

Distance(A, B) = A XOR B

This creates a metric space where:

• Each node knows about nodes "close" to it

• Lookups get exponentially closer with each hop

Routing Table

Each node maintains a routing table of k-buckets:

┌─────────────────────────────────┐
│ Bucket 0: Distance 2^0 - 2^1    │  ← Very close nodes
├─────────────────────────────────┤
│ Bucket 1: Distance 2^1 - 2^2    │
├─────────────────────────────────┤
│ Bucket 2: Distance 2^2 - 2^3    │
├─────────────────────────────────┤
│ ...                             │
├─────────────────────────────────┤
│ Bucket 255: Distance 2^255+     │  ← Very far nodes
└─────────────────────────────────┘

Each bucket holds up to K nodes (K=20 in KayakNet).

Operations

PING

Check if a node is alive:

Node A → PING → Node B
Node A ← PONG ← Node B

STORE

Store a key-value pair:

Node A → STORE(key, value) → Closest nodes

FIND_NODE

Find nodes close to a target ID:

Node A → FIND_NODE(target_id) → Node B
Node A ← [list of close nodes] ← Node B

FIND_VALUE

Retrieve a stored value:

Node A → FIND_VALUE(key) → Node B
Node A ← value OR [closer nodes] ← Node B

Peer Discovery Process

Joining the Network

1. Connect to bootstrap
   │
2. Get initial peers
   │
3. Perform FIND_NODE on own ID
   │
4. Discover close neighbors
   │
5. Populate routing table
   │
6. Announce presence

Iterative Lookup

To find a node or value:

Start with α closest known nodes
    │
For each node, send query
    │
Receive responses with closer nodes
    │
Query the newly discovered closer nodes
    │
Repeat until no closer nodes found
    │
Return result

Parameters:

• α (alpha) = 3 (parallel queries)

• K = 20 (nodes per bucket)

Data Storage

What's Stored in DHT

Key Type	Value	TTL
peer:<id>	Node address	1 hour
room:<name>	Room metadata	24 hours
domain:<name>	Domain record	Permanent
listing:<id>	Marketplace listing	7 days
msg:<hash>	Message (store-forward)	1 hour

Replication

Data is stored on the K closest nodes:

• Provides redundancy

• Survives node failures

• Automatic re-replication

TTL and Refresh

• Records have TTL

• Publishers refresh before expiry

• Expired records are deleted

• Popular content stays available

Configuration

{
  "dht": {
    "k": 20,              // Bucket size
    "alpha": 3,           // Parallel queries
    "record_ttl": "24h",  // Default TTL
    "refresh_interval": "1h"
  }
}

Performance

Lookup Efficiency

For N nodes in network:

• Average hops: O(log N)

• With 1M nodes: ~20 hops

• With caching: Often 1-3 hops

Bandwidth Usage

• PING: ~100 bytes

• FIND_NODE: ~500 bytes

• STORE: Variable (depends on value)

Security Considerations

Sybil Protection

Without protection, attacker could:

• Create many nodes

• Control routing

• Censor lookups

KayakNet mitigation:

• Proof-of-Work for new nodes

• Peer scoring

• Diverse peer selection

Eclipse Protection

Without protection, attacker could:

• Surround victim node

• Control all their peers

• Censor or manipulate

KayakNet mitigation:

• Require peers from different subnets

• Maintain bootstrap connections

• Periodic peer refresh

Data Integrity

All DHT records are signed:

Record = {
  key: "domain:example.kyk",
  value: { ... },
  publisher: <public_key>,
  signature: <Ed25519_sig>,
  timestamp: 1234567890
}

Unsigned or invalid records are rejected.

Comparison with Other DHTs

DHT	Used By	Differences
Kademlia	BitTorrent, KayakNet	XOR distance, k-buckets
Chord	Academic	Ring topology
Pastry	Academic	Prefix routing
Coral	CDN	Hierarchical clusters

Troubleshooting

Few Peers

# Check peer count
curl http://127.0.0.1:8080/api/network/peers

# If < 5, wait or restart

Slow Lookups

• Check network connectivity

• Verify bootstrap is reachable

• Increase alpha for more parallelism

Data Not Found

• Publisher may be offline

• TTL may have expired

• Try again after network refresh