Overview
At Sireto Technology, the fintech platform needed a way for network nodes to discover each other without relying on a central registry — a single point of failure that would be unacceptable in a financial transaction network. I implemented the Kademlia Distributed Hash Table (DHT) protocol in Java to solve this.
Kademlia is the algorithm underlying BitTorrent’s distributed tracker and Ethereum’s devp2p node discovery. It provides O(log n) peer lookup with high fault tolerance — nodes can join and leave freely, and the network heals itself.
How Kademlia Works
Kademlia organizes nodes in a virtual XOR-metric space. Each node has a 160-bit ID; distance between nodes is computed as the XOR of their IDs. Each node maintains a routing table of k-buckets — lists of known peers at exponentially increasing distances.
To look up a peer, a node iteratively queries the closest known nodes to the target ID. Because of the XOR metric, each step halves the distance — lookup converges in O(log n) hops.
What I Implemented
- Node bootstrap — joining the network by contacting a known seed node, then populating the routing table via iterative
FIND_NODERPCs - k-bucket management — maintaining and refreshing buckets, evicting stale nodes, handling the “least-recently seen” eviction policy Kademlia uses for reliability
- Core RPCs over UDP:
PING,STORE,FIND_NODE,FIND_VALUE - Iterative lookups — alpha-parallel async lookup with per-round closest-set tracking
- Network churn handling — periodic bucket refresh and republish to keep routing tables accurate as nodes come and go
Integration
The Kademlia layer was integrated into the broader fintech platform as the peer discovery backbone — nodes used it to find transaction validators and relay peers without a central broker.
Stack
Java · UDP Sockets · Kademlia DHT · Distributed Systems · Concurrent Programming