QUICompare
Started in the second half of 2025, QUICompare carries the same research line forward: it measures HTTP/1.1, HTTP/2 and HTTP/3 (QUIC) traffic on Apache and Nginx servers under different TCP congestion-control algorithms, and natively computes the same fractal metrics — Hurst, fractal dimension, alpha-tail — that the WMDEA, SMDEA and Super-Cobb-Douglas models use as decision variables.
From controlled load to fractal metrics
What it does
What it compares
Why not ApacheBench, isolated curl, or h3load?
Classic tools — — are useful for one-off cases, but don't simultaneously solve:
Why TCP, UDP and the kernel still matter
RFC 2544 and RFC 6815 — discipline, not overload
QUICompare engages with RFC 2544, but doesn't limit itself to reproducing it: it uses it as a reference for experimental discipline. RFC 6815 reinforces that these methods shouldn't be applied directly to production networks — only in isolated lab environments.
| Reference | Role in QUICompare | Practical implication |
|---|
Nine commands, four families
Real output of quicompare --help on v4.7.3 — valid across every
recent version of the tool. Descriptions marked inferred
don't come from documentation; they're read off the command name and the tool's
own tagline ("Mono & Multi fractals Analysis"), and may be incomplete.
Congestion-control algorithms supported in --auto-research
Real command examples (official tutorial)
What needs to be installed
From the .deb package to the first test
Install the package
sudo dpkg -i quicompare_4.7.3_amd64.deb — from a ready release or by building build-quicompare-deb.sh.
Automatic setup
Creates the Python virtualenv, generates an ed25519 SSH key if missing, and compiles a curl with HTTP/3 support (quictls + ngtcp2 + nghttp3) — 10 to 20 minutes, one time only.
Authorize the DUT
quicompare-import-ssh-keys --dut-host <IP> --dut-user root — only the public key leaves the client machine.
Validate
quicompare --version and curl --version should list HTTP3 in the features before the first real test.
Two VMs: client and DUT
The tutorial's base scenario uses two lab VMs, available for download as OVA files.
Why curl-http3 + ffmpeg
Comparing HTTP/2 with one load tool and HTTP/3 with another, with very different behavior (h3load, quiche), turns the stress generator itself into a confounding variable. QUICompare stabilizes this by keeping the same file and the same consumption pattern on both sides.
| Layer | HTTP/2 | HTTP/3 |
|---|
Experimental dataset
Download the dataset used as a reference for the QUICompare experimental analysis.
Dataset disponibilizado no Mendeley Data.
One run's report, unedited
Below is a real run summary (--superlight), generated on 2026-08-04.
It's a short debug test — the tool itself flags this in the warnings below.
View the full JSON, as the tool generated it
Exported variables, by group
| Group | Variables |
|---|
From QUICompare back to multicriteria decision-making
In publication-grade runs (≥120 s, ≥10 clients, repeated), QUICompare computes Hurst, fractal dimension and alpha-tail classification for every combination of server, HTTP version and congestion algorithm — exactly the decision variables that the WMDEA, SMDEA and Super-Cobb-Douglas models use to rank configurations. QUICompare still has no dedicated publication; according to the official tutorial, the resulting decision matrix is meant to feed a broader family of methods: