Benchmarking Nextcloud

What is Nextcloud? A safe home for all your data. Access & share your files, calendars, contacts, mail & more from any device, on your terms.

Most of the Nextcloud is based of the code supplied in https://github.com/nextcloud/docker please refer to this documentation for everything Nextcloud dependent.

We evaluate the two browsers Chromium and Firefox. Also the three databases MariaDB, Postgres and SQLite.

Install nextcloud beforehand with defaults

Please note that this scenario should be run with 3600s idle and 3600s baseline durations

python3 runner.py --uri ~/Sites/green-coding/nextcloud-energy-tests --filename usage_scenario-mariadb-install-firefox.yml --dev-no-optimizations --allow-unsafe --name "Nextcloud Blue Angel - Install"

This will install nextcloud with the defaull apps. Currently:

• Calendar
• Contacts
• Mail
• Document editor
• Talk (Video conference & Instant Messaging)

All changes are saved in the docker volumes for Apache webserver and the MariaDB database.

Preparation of installed Nextcloud

Nextcloud by default has a cronjob that will trigger in parallel with web requests.

This must be deactived in the Admin Dashboard under Basic Settings. Setting it to cron will effectively deactivate it if no cron job is set to trigger the .php file afterwards.

Execution of the tests for the Blue Angel certification

For a proper run login via SSH to the machine and then start the process unattended.

First prepare the system to allow firefox window to open from inside docker

export DISPLAY=:0
xhost +local:docker # might also need to be done on the machine directly. Not only via ssh

Also configure GMT to use 60s pre-test sleep time. Idle should be at 120 s to make container energy estimation work. Baseline can be at 0.

Then execute the scenarios via GMT

nohup bash run-tests.sh &> output.log &
disown -h %1
logout

Scenarios details

Currently there are six different usage scenarios of Nextcloud, orchestrated with different databases and browsers. All scripts will start of with creating an admin account and installing the recommended apps,
then each case diverges from that point. The cases are:

• Contacts: Nextcloud login via the web interface. Create a new contact entry. Edit the contact entry. Delete the contact entry.
• Files: Nextcloud login via the web interface. Create a new file. Create a sharing link for the file with 1 MB. Share the file and download it with another browser.
• Calendar: Nextcloud login via the web interface. Create a new calendar entry. Edit the calendar entry. Delete the calendar entry.
• Video: Video conference with Nextcloud Talk app. First, a call is started from a browser and a sharing link is created. Then two participants join the call. The video conference is simulated via Firefox with a color-changing demo video and noise as audio.
• Docs: Another user is created. Then an empty Markdown document is created and shared with this user. Both users then access the document simultaneously and edit it collaboratively. Each user takes turns adding 50 characters to the document and it is checked that this also reaches the other users. The input is simulated with a 200ms delay per character.
• Talk: Send text messages with the Talk app from Nextcloud. A conversation is created by a user and then 5 other users are added via a link. They all send separate messages with a length of 50 characters. It is validated that the messages reach all parties. Input is simulated with a 200ms delay between keystrokes.

Infrastructure - Apache / MariaDB

The compose file is copy pasted from https://github.com/nextcloud/docker#base-version---apache We set a bogus password for *_ROOT_PASSWORD and *_PASSWORD as the GMT requires environment variables to be set.

Running

You will need to supply the --allow-unsafe to the runner as it will mount volumes.

Debugging tips

In cases where the script hangs because of a wrong locator, it might be helpful to run the usage scenario in debug mode and then attach a shell to the Playwright container. Running playwright codegen -o output.py from the container and then following the actions of the script can help investigate what exactly is different.