The probes show what happens on real networks, where many things change at once. Experiments change one thing on purpose — packet loss, delay, a blocked protocol — between two servers we control, and measure what each VPN protocol still does. Every experiment publishes its scripts and raw results so it can be repeated.
Four VPN protocols between two servers while the network drops 0, 1, 5 and 10 % of packets in both directions. Download speed and page-load times, three runs each.
No VPNreferenceHysteria2VLESS RealityWireGuardOpenVPN (UDP)
Axis 0–100 %Loss levels: categories
Data table
Speed kept
0 % loss
1 % loss
5 % loss
10 % loss
No VPN
100%
97%
88%
77%
Hysteria2
100%
96%
91%
47%
VLESS Reality
100%
97%
90%
78%
WireGuard
100%
48%
15%
4%
OpenVPN (UDP)
100%
28%
12%
4%
Planned
Planned · no results yet
UDP blocked
The same protocols on a network that drops all UDP, as some networks do wholesale: which transports still connect, and how quickly an app falls back to one that does.
Planned · no results yet
High latency (1000 ms)
A very slow round trip, as on satellite links or badly congested routes: handshakes, page loads and download speed per protocol.
Planned · no results yet
Throttled LTE profile
A mobile network that throttles bandwidth and varies delay and loss over time, closer to a real congested cell than fixed settings.
Scripts and raw data
All experiment scripts, configurations and raw results are available on GitHub.
Experiments run between controlled servers and change one network variable at a time. They are separate from the continuous measurements; they do not feed the success rates on the other pages.
Reproducibility principles
Scripts and configurations are published
Raw results are published
Experiments can be repeated by others
Setup and parameters are documented on each experiment page