ringkernel-txmon

GPU-accelerated real-time transaction monitoring showcase for RingKernel.

Overview

TxMon demonstrates RingKernel's capabilities for high-throughput financial transaction monitoring, suitable for banking/AML compliance scenarios. It includes a transaction factory, compliance rule engine, and real-time GUI.

Features

• Transaction Factory: Configurable synthetic transaction generation with realistic patterns
• Compliance Rules: Velocity breach, amount threshold, structuring detection, geographic anomaly
• Real-time GUI: Live transaction feed, alerts panel, statistics dashboard
• Multi-backend: CPU and optional CUDA GPU acceleration

GPU Backend Approaches

The cuda module provides three different GPU-accelerated approaches:

1. Batch Kernel (High Throughput)

Each thread processes one transaction independently. Best for maximum throughput on large batches.

// ~93B elem/sec on RTX Ada
let alerts = batch_kernel.process(&transactions).await?;

2. Ring Kernel (Actor-based Streaming)

Persistent GPU kernels with HLC timestamps and K2K messaging. Best for low-latency streaming with complex multi-stage pipelines.

let kernel = ring_kernel.launch("monitor", config).await?;
kernel.send(transaction).await?;

3. Stencil Kernel (Pattern Detection)

Uses GridPos for spatial pattern detection in transaction networks. Detects circular trading, velocity anomalies, and coordinated activity patterns.

let config = StencilConfig::new("pattern_detect")
    .with_tile_size(16, 16);
let patterns = stencil_kernel.detect(&network).await?;

Usage

use ringkernel_txmon::{TransactionGenerator, MonitoringEngine, GeneratorConfig};

// Create transaction generator
let config = GeneratorConfig {
    transactions_per_second: 10000,
    fraud_ratio: 0.01,  // 1% fraudulent
    ..Default::default()
};
let generator = TransactionGenerator::new(config);

// Create monitoring engine
let engine = MonitoringEngine::new(MonitoringConfig::default());

// Process transactions
for tx in generator.generate_batch(1000) {
    if let Some(alert) = engine.evaluate(&tx) {
        println!("Alert: {:?}", alert);
    }
}

Alert Types

Run GUI

cargo run -p ringkernel-txmon --bin txmon

Benchmark

cargo run -p ringkernel-txmon --bin txmon-benchmark --release --features cuda-codegen

Typical results on RTX 4090:

Testing

cargo test -p ringkernel-txmon

The crate includes 40+ tests covering transaction generation, rule evaluation, and GPU kernels.

License

Apache-2.0