QUICK- Performance Benchmarks

TCP-over-QUIC tunnel optimized for satellite networks (LEO/MEO/GEO).

These benchmarks run automatically after each merge to main and track performance regressions over time.

How to Read These Charts

X-axis = commit (short SHA). Each point is a merge to main, left-to-right = oldest to newest.
Y-axis = measured value. Each chart states its unit and whether lower or higher is better.
Trend going up = the value is increasing over time. Good or bad depends on the metric (read the per-chart label).
Click any data point to open the exact commit on GitHub that produced that measurement.
Hover over a point to see the commit message, timestamp, exact value, and variance range.
Flat line = stable performance. Sudden spike = potential regression (or improvement) to investigate.

Microbenchmarks (Criterion)

Unit-level performance measurements using Criterion.rs. Each benchmark runs thousands of iterations and reports the average time per iteration in nanoseconds (ns). Lower is better — it means the operation completes faster.

The variance range in tooltips shows statistical spread. A tight range means consistent performance; a wide range suggests noise (CPU contention, etc.).

System Benchmarks (E2E Throughput)

End-to-end tunnel throughput: real TCP traffic sent through the full QUICK- QUIC tunnel under simulated network conditions using Linux tc netem (delay, bandwidth cap, packet loss). Measured with iperf3 over 30-second sessions.

Profile	RTT	Downlink	Uplink	Loss	Simulates
Terrestrial	10 ms	100 Mbps	100 Mbps	0%	Fiber/LAN baseline
LEO	50 ms	50 Mbps	10 Mbps	0.1%	Starlink-class
MEO	250 ms	30 Mbps	5 Mbps	0.5%	O3b/SES-class
GEO	600 ms	10 Mbps	2 Mbps	1%	Eutelsat-class

The Downlink/Uplink columns show the link capacity configured in tc tbf. Actual tunnel throughput will be lower due to protocol overhead, congestion control ramp-up, and packet loss recovery.