Decentralized Peer-to-Peer Load-Balancing System

A lightweight C++/gRPC demo that implements a peer-to-peer, decentralized load-balancing and failover algorithm. Each node periodically heartbeats its own “score” (40% CPU, 30% queue depth, 20% memory‐free, 10% uptime plus jitter) to its neighbors. Incoming tasks are routed to the highest-scoring alive node; on failure, they automatically retry down the score list.

---

Features

• Weighted scoring based on CPU load, queue length, memory usage, and uptime.
• Heartbeat gossip: each node learns peer scores (and -hops) up to configurable depth.
• Fault tolerance: tasks forward to the next best node if the best one fails, with client and server retry logic.
• Metrics & logs: per-node JSONL metrics, detailed event logs.

---

Prerequisites

• C++17 compiler
• gRPC & Protobuf (v1.35+ recommended)
• CMake (v3.12+)
• Unix-like OS (Linux, macOS)

---

Build

chmod +x start.sh
./start.sh
mkdir -p build && cd build
cmake ..
make -j

---

Running

Start servers like this:

In build folder run

./server <address of your device>:<port number> <address of your neighbour>:<port number> ...

Start client like this:

./client <TASK_ID> "<PAYLOAD>" <REPLY_ADDR> <SERVER1> [<SERVER2> ...]

Configuration

• Score weights: Adjust the constants in Server::computeScore().
• Max gossip hops: Currently hard-coded to 3, can be made configurable.
• Heartbeat interval & timeout: Edit heartbeatLoop() and gRPC deadlines in server.cpp.

Logging & Metrics

• Event logs: JSON lines of named events for debugging (logs/*.log).
• Metrics: JSON lines recording CPU%, memory%, uptime, and score per second (logs/*.metrics.jsonl).