Measured 7 September 2026 · best of five runs

The QNet Protocol,
measured.

Two devices, connected directly. No server in between.

What the QNet Protocol changes
YouPeerServersees who talks to whom
The usual way
YouPeernoneencrypted, direct
QNet Protocol

Normally your traffic passes through a company's server. Even when the content is encrypted, that server still sees who talks to whom, and how often. QNet connects the two devices directly, so there is no such record, because there is nowhere to write it down.

Sovereignty

No middleman means no dependency on someone else's prices, terms, outages or jurisdiction.

Privacy

Who you talk to often reveals more than what you say. Direct connections leave no such trail.

Quantum-safe

Traffic recorded today can be decrypted later, once quantum computers mature. QNet can add a quantum-safe key exchange (ML-KEM-768) to the connection setup.

Why quantum-safe, today

Recorded now,
opened later.

An attacker does not need a quantum computer today. They only need to store your encrypted traffic until they have one. Anything with a long shelf life (contracts, health records, source code, state communication) is therefore already at risk now.

The defence has to be in place before that day, not after it. Adding it later does not help: the recording already happened.

TodayTrafficrecordedstored, waitingQuantum eradecryptedWith ML-KEM-768: still unreadable
Head to head

How we compare

Three questions decide whether a protocol is usable in practice: what protection costs, how often you still need someone else's server, and how long it takes to connect.

Cost of protection

Encryption costs us under 8 %, while doing more than the others.

Every protocol adds overhead on top of raw, unprotected data. WireGuard is leaner because it runs inside the operating system kernel and does far less: no peer discovery, no NAT traversal. QNet does all of that in userspace and still stays close.

Lower is better · ¹ published sources · ² measured
0%5%10%15%20%25%30%35%40%QNet Protocolunder 8 %Finds peers, punches through routers andencrypts: all in normal software, for under 8%.WireGuard~5 %Leaner because it lives inside the operatingsystem, but it cannot find peers or crossrouters on its own.ZeroTier~17 %Twice our cost, and about a third of itsconnections still go through its own servers.libp2p~22 %A toolkit rather than a finished protocol:flexible, but the layers add up.WebRTC~30 %Built for browser video calls; the widestreach, at the highest cost per packet.Raw UDP0 % (baseline)Plain data with no protection at all. Theyardstick everything else is measured against.

Independence from third-party servers

We never fall back to a relay. The others do, for up to a third of all sessions.

This is the number that decides sovereignty. A relay sees the metadata, costs money per gigabyte, and is a single point that can fail or be shut down. For QNet it is not a rare fallback: there is none.

Share of sessions routed through a third party · lower is better
0%5%10%15%20%25%30%35%40%QNet ProtocolneverNo fallback server exists, so none can readmetadata, bill you, or be shut down.WebRTC15-20 %When a router refuses, traffic is relayedthrough a TURN server you have to run and payfor.ZeroTier~30 %Roughly one in three sessions is routedthrough ZeroTier infrastructure.
Deep dive

Inside the protocol

Four measurements, taken separately so it stays visible where the time actually goes.

Cryptography

11.3M
key rotations / hour

Connection setup

0.01
milliseconds

Direct connections

11,158
pairs / second

Reconnect

0.31
milliseconds

Encryption throughput

A fresh key per message: 11.3 million times an hour.

Most systems keep one key for minutes at a time. If it leaks, everything sent in that window is readable. QNet replaces the key constantly, so a stolen key unlocks a single message. That is only possible because a full cycle costs 0.29 milliseconds.

Operations per second · higher is better
05k10k15k20kEncrypt14,445/sEncrypting a message takes 0.07 ms, soprotection never becomes the bottleneck.Key generation8,357/sA brand-new key pair every 0.05 ms, which iswhat makes constant key changes affordable.Full rotation3,137/sGenerate, encrypt, decrypt: the whole cycle in0.29 ms, over 11 million times an hour.
Response time per operation

Averages hide the bad cases. Here even the slowest 1 in 100 stays under half a millisecond: the speed holds up under load, it is not just a flattering average.

0.000.120.230.350.460.05 ms0.28 ms0.39 msKey generation0.07 ms0.08 ms0.09 msEncryption0.29 ms0.42 ms0.46 msFull rotation
Typical95 out of 10099 out of 100

Recovery after an IP change

Switching network does not drop the session.

Walking out of the office moves a phone from Wi-Fi to mobile and changes its address. Most systems rebuild the whole connection. QNet restores the existing session instead, fast enough that a call or a transfer keeps running.

Milliseconds · lower is better
0 ms0.3 ms0.5 ms0.8 ms1 msTypical0.31 msThe usual case: your session is back beforeyou notice the network changed.95 out of 1000.71 ms19 out of 20 reconnects finish faster thanthis.99 out of 1000.76 msEven the worst case stays under a millisecond,with no dropped calls or transfers.

Scaling with hardware

Twice the hardware, nearly twice the throughput.

Many systems stop improving once cores are added, because the parts start waiting for each other. QNet still reaches 82 % of perfect scaling at sixteen parallel workers: capacity can be bought with hardware instead of redesigned.

82 %of perfect scaling
at 16 workers
018k35k53k71kN=1N=2N=4N=8N=16N=1Measured: 4kPerfect scaling: 4kN=2Measured: 9kPerfect scaling: 9kN=4Measured: 18kPerfect scaling: 18kN=8Measured: 36kPerfect scaling: 35kN=16Measured: 58kPerfect scaling: 71kpairs / second
MeasuredPerfect scaling

Twelve hours without a stumble

Fast for a second is easy. We ran it for half a day.

A benchmark measures seconds. Real deployments run for months, and that is where slow leaks and rare races surface. So we left a fifteen-node network running unattended for twelve hours and took a sample every fifteen minutes. All 49 samples found every node connected to every other one: no restarts, no failed decryptions, no silent degradation.

Duration
12 hours, unattended
Network
15 nodes, full mesh
Samples
49, every 15 minutes
Mesh alive
49 of 49 samples at 100 %
Restarts
0
Decryption failures
0

A separate three-hour run with the quantum-safe key exchange switched on produced the same result: full mesh, no restarts, no decryption errors.

How we measured

Machine
12 dedicated cores, 55 GB memory
Throughput run
5 runs × 10 s per dimension
Endurance run
12 h, 15 nodes, 49 samples
Date
Throughput 7 Sep 2026 · endurance Jul 2026
Setup
Local connections, no internet in between
Not included
Internet travel time

No tracking cookies. The partner data room sets a session cookie. Details