Articles

These articles are raw notes on design decisions that surface while hacking on gtp-guard. Each one digs into a specific problem hit during design, implementation, or testing, and documents why a particular approach was picked over the alternatives. They reflect where the project is heading technically.

5G UPF: Smart Session Placement · 2026-04-19 · 5G
Goes beyond single-metric scheduling with trend-based algorithms that track load trajectory, a Weighted Score Composite that blends sessions, load, bandwidth, and PPS into one score, and a Constraint-Based Scheduler that enforces hard limits before delegating to any other algorithm.

5G UPF: From Flow Steering to Session Affinity · 2026-04-17 · 5G
Integrates range partitioning into the PFCP control plane so every session establishment becomes a scheduling decision the NIC enforces at line rate. Introduces range-partition, flow-steering-policy, and cpu-sched-group objects, and shows how they align TEID, IP pool, and CPU for the entire session lifetime.

Per-Core CPU Load Measurement · 2026-04-02 · System
Explains why /proc/stat, /proc/schedstat, and task-clock counters miss inline XDP/NAPI softirq work, and why GTP-Guard drives scheduling from PERF_COUNT_HW_REF_CPU_CYCLES to get accurate per-core load even when packets are processed outside the scheduler's view.

5G UPF: Pump Up the Volume! · 2026-03-27 · 5G
Lays the hardware foundation for a 800 Gbps UPF on a 1RU commodity server: NIC rx_queue pinning, IRQ affinity, range partitioning math on TEID and IP sub-prefixes, and end-to-end flow steering validation with TC flower rules.