Smart Network & RPC Observability (eBPF + ML + Adaptive)

Draft implementation for optimizing network performance and diagnosing latency, packet loss, jitter, and bottlenecks for high-load blockchain nodes and RPC infrastructure. Architecture: Observe → Reason → Act.

• Paper: WHITEPAPER.md
• Folder: Self-contained in rpc-network-intelligence/ (sibling to delphi-showcase-project and interview-prep).

Architecture

Stage	Component	Role
Observe	src/collector.py	Produces eBPF-like metrics (RTT, loss, jitter, RPC, node). Demo uses simulated data; production would use eBPF.
Reason	src/features.py, src/model.py	Feature extraction, anomaly detection (Isolation Forest), root-cause classification (network / node / RPC).
Act	src/adaptive.py	Turns ML output into suggested actions and alerts (no automatic routing changes).

Setup

cd rpc-network-intelligence
python3 -m venv .venv
source .venv/bin/activate   # or .venv\Scripts\activate on Windows
pip install -r requirements.txt

Run pipeline

From the rpc-network-intelligence directory:

python run_pipeline.py

This will:

1. Build synthetic training data (normal + network/node/rpc anomaly windows).
2. Fit the anomaly detector and root-cause classifier.
3. Run four demo windows (normal, network-bound, node-bound, rpc-bound) and print adaptive suggestions.

Config

• Metrics schema: config/metrics_schema.yaml — aligns with the white paper §3.

Extending

• Real eBPF: Replace collector.collect_window() input with metrics from eBPF programs (e.g. BCC/libbpf) or from Prometheus/OpenTelemetry exporters that follow the same schema.
• Dashboard: Export metrics and ML outputs to Prometheus/Grafana or OpenTelemetry.
• Automation: In production, the adaptive layer can call internal APIs (load balancer, BGP policy) with safeguards.

License

Same as the parent repository.