Network QoE Measurement Framework

All code, data and plots for the master's thesis Characterizing Application Behavior and Quality of Experience (QoE) in 5G Networks (Adrian Hovatn Marstokk, NTNU).

This README covers how the repo is laid out, how to reproduce the figures, and how to run the testbed.

Layout

Path	Contents
`playbooks/main.yml`	Imports the 8-stage pipeline
`playbooks/pipeline/0..7`	reset → init → set CC → server up → run client → server down → check → reset CC
`playbooks/sanity/`	Tests and verification
`scenarios/`	Experiment and test scenarios: `phase1/{RQ1a,RQ1b,RQ1c}`, `phase2/5g/`, `sanity/`
`scripts/run_client_apps.py`, `scripts/setup_ns.sh`	Client-side app launcher + per-app network-namespace setup
`scripts/run_experiments/`	Batch runners + pre-experiment run checks/tests
`scripts/scenarios/`	Scenario generators (`gen_phase1.py`, `gen_phase2.py`)
`scripts/post_run_data_processing/`	`extract_data.py` + archive helpers
`scripts/analysis_and_plotting/`	`plot_*.py`: one per thesis figure family
`scripts/sanity_checks/`	All tests, more information in sanity_checks/README
`scripts/testbed/`	Standalone utilities
`inventory/`	Ansible inventory + group vars
`diagrams/`	Architecture diagrams (mermaid source + PNGs)
`data/`	Thesis dataset (Phases 0–2)
`output/`	Per-batch runner logs
`plots/`	Committed figures, read Reproduce the figures for thesis figure to file mapping
`sanity/`	Preflight check logs (gitignored) + 5G speedtest measurements

Thesis phase → repo path

Directory names predate the thesis phase numbering, so there's no phase0/, and repo scenarios/phase1/ holds thesis Phase 0 and 1. The table maps each phase:

Thesis phase	Scenarios	Run as	Plots
Phase 0	`scenarios/phase1/`	LAN (default)	`plots/LAN/`
Phase 1	`scenarios/phase1/` (same files)	`-e transport=5g`	`plots/5g/phase1/`, `plots/comparison/LAN_vs_P1_vs_P2/`
Phase 2	`scenarios/phase2/5g/`	`-e transport=5g`	`plots/5g/phase2/`, `plots/comparison/{p5a_vs_p5b,p6_compare_25-50-75,P3-P6_compare}/`
Phase 3	(analysis only)	—	`plots/comparison/phase3_*.png`

Reproduce the figures

No testbed required, only requirements.txt and python:

pip install -r requirements.txt
python3 scripts/analysis_and_plotting/plot_cdf_phase1.py   # one plot_*.py per figure family

Each plot_*.py reads data/ and writes plots/.

Thesis figure → file

plots/ mirrors data/. Cited figures:

Thesis figure	Repo path
LAN bandwidth / RTT-impact CDFs	`plots/LAN/CUBIC/cdf_{bw,rtt}_impact_*`
LAN client-count CDFs	`plots/LAN/UE_impact/cdf_ue_impact_*_cubic`
CUBIC vs BBR vs RENO CDFs	`plots/comparison/cubic_vs_bbr_vs_reno/`
QoE per CC	`plots/comparison/phase1_qoe_by_cc.png`
VoIP invariance	`plots/comparison/phase1_voip_invariance.png`
LAN vs P1 vs P2 (`P<T` bars)	`plots/comparison/LAN_vs_P1_vs_P2/`
P3 / P4 bandwidth-impact CDFs	`plots/5g/phase2/P{3,4}/CUBIC/`
P5a vs P5b CDFs	`plots/comparison/p5a_vs_p5b/`
P6 reservation-level CDFs	`plots/comparison/p6_compare_25-50-75/`
7-policy `P<T` bars	`plots/comparison/P3-P6_compare/`
Per-policy QoE / utilization	`plots/comparison/phase3_{qoe,util}_by_policy.png`

Everything else plot_*.py outputs is supplementary and not cited.

Run the testbed

Running experiments needs the physical testbed: sp-client, sp-server, sp-proxy (orchestrator). For 5G, the RUTX50 UEs, srsRAN gNB, and Open5GS core are needed. Hostnames resolve only on the orchestrator (its /etc/hosts + ~/.ssh/config), so these commands won't run from a fresh clone elsewhere.

Inventory host	Group	Network	Address
sp-proxy (orchestrator)	`[orchestrator]`	lab LAN + tailnet	`10.100.52.61` (tailnet `100.94.217.61`)
sp-client	`[client]`	tailnet	`100.126.60.53` (wired NIC is DHCP)
sp-server	`[server]`	lab LAN + tailnet	`10.100.52.62` (tailnet `100.69.251.59`)
5gs	`[upf]`	lab LAN	`10.100.52.15` — Open5GS UPF + `ogstun`
gnb	`[gnb]`	lab LAN	`10.100.52.17` — srsRAN gNB

Tailscale was added for the 5G phases and is how the inventory addresses hosts now: sp-client's wired IP is DHCP-assigned, so the tailnet gives a stable address to SSH into and orchestrate it. Phase 0 (LAN) predates Tailscale and ran over the wired 10.100.52.0/24 LAN. For backwards-compatibility, the testbed now routes LAN runs over the tailnet too, so a true original Phase 0 rerun needs the wired path.

Testbed prerequisites

setup.sh installs only the orchestrator's Python/Ansible toolchain. The traffic-generating tools are not bundled or installed by any playbook

Host	Software	Used for
sp-client	goDASH (path set in `inventory/group_vars/all.yml` → `godash_bin`)	DASH video client
sp-client	D-ITG sender `ITGSend`	VoIP traffic
sp-client	`curl`	file downloads
sp-server	Apache (`apache2`) on `:80`, serving the DASH content at `mpd_url` + a writable `filedl_dir`	video + file download server
sp-server	D-ITG `ITGRecv` + `ITGDec`	VoIP receive + decode

The Tears of Steel DASH dataset (referenced by mpd_url) must be placed under Apache's document root to match that URL. The file download payloads under filedl_dir are generated by the server playbook. iperf3 is only needed for the optional scripts/testbed/*_speed_test.sh.

First-time setup on the orchestrator:

./setup.sh

setup.sh installs the Python deps + Ansible + collections, creates .env from .env.example (never overwriting an existing one), and checks SSH to the testbed hosts. On externally-managed Python (Ubuntu 24.04+, PEP 668) run it in a venv first (python3 -m venv .venv && source .venv/bin/activate). Fill .env with the real secrets (sudo + RUTX50 passwords). scripts/sudo_pass.sh reads SUDO_PASSWORD from it for the --become-password-file in the commands below.

Single scenario:

# LAN (default)
ansible-playbook -i inventory/hosts.ini playbooks/main.yml \
  -e @scenarios/phase1/RQ1a/20mbits-30rtt-cubic.yml \
  --become-password-file scripts/sudo_pass.sh

# 5G: add the transport + server-IP overrides
ansible-playbook -i inventory/hosts.ini playbooks/main.yml \
  -e @scenarios/phase1/RQ1a/20mbits-30rtt-cubic.yml \
  -e transport=5g -e server_ip=10.100.52.62 \
  --become-password-file scripts/sudo_pass.sh

Pre-run checks:

./scripts/preflight_checks.sh                          # tests/sanity checks
./scripts/run_experiments/LAN_verify_before_runs.sh    # LAN preflight checks
./scripts/run_experiments/5g_verify_before_runs.sh     # 5G preflight checks

Batch runs (default 5 reps, REPS=N env override):

./scripts/run_experiments/run_scenarios.sh -d scenarios/phase1/RQ1a -n 5  # generic
./scripts/run_experiments/LAN_run_phase1.sh              # Phase 0
./scripts/run_experiments/5g_run_phase1_shared_slice.sh  # Phase 1, P1
./scripts/run_experiments/5g_run_phase1_per_pdu.sh       # Phase 1, P2
./scripts/run_experiments/5g_run_phase2_p3.sh            # Phase 2 P3
./scripts/run_experiments/5g_run_phase2_p4.sh            # Phase 2 P4
./scripts/run_experiments/5g_run_phase2_p5.sh            # Phase 2 P5
./scripts/run_experiments/5g_run_phase2_p6.sh 25         # Phase 2 P6 — floor variant: 25 | 50 | 75

Runs land raw archives in ~/exp_results/ (outside the repo), extract a batch into data/, then plot via Reproduce the figures:

python3 scripts/post_run_data_processing/extract_data.py --batch-dir output/<...>/batch_<ts>

Citing this testbed

Bibtex to cite this testbed:

@misc{marstokk2026qoetestbed,
  author       = {Marstokk, Adrian Hovatn},
  title        = {2026-thesis-adrianhm},
  howpublished = {\url{https://git.ntnu.no/ie-iik/2026-thesis-adrianhm}},
  year         = {2026},
  note         = {Testbed code for the master's thesis: 
                  ``Characterizing Application Behavior and Quality of Experience (QoE) in 5G Networks'',
                  Norwegian University of Science and Technology (NTNU)}
}

License

MIT — see LICENSE.

License

ie-iik/2026-thesis-adrianhm

About

Resources

License

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages