What it costs, measured.

The machine and the versions

The cost of a stack

Numbers 1 and 2 come from the same stack: seventeen Laravel projects in the projects directory, one app container, one mariadb container.

Images, as docker reports them

Size on disk, arm64. The app image carries nginx, php-fpm, composer, node, npm, yarn, phpunit, xdebug and the mysql and postgres clients, so there is nothing else to install for a project.

Two of these are worth a sentence: mariadb is the lighter SQL server (310 MB against 795 MB for mysql 8.0, both with the data directory empty), and elasticsearch is the one heavy profile — everything else in the stack fits in a few hundred megabytes.

Idle memory

docker stats --no-stream, after the stack was up and then left alone for 60 s:

So the default install — app and mariadb, serving every project — idles at about 113 MiB. Add redis and memcached for under 8 MiB more. MySQL 8.0 costs roughly five times what MariaDB costs in memory (375 MiB against 78 MiB, same configuration), which is worth knowing before you pick the mysql profile over mariadb. Elasticsearch is the exception in every stack, and here it is the only container that costs gigabytes: 6.4 GiB in docker stats, with the JVM and the memory-mapped index files counted against it, and tens of seconds of start-up.

Time to start

These are warm numbers: the images are pulled and the data volumes exist. A first run also initializes the data directory of a database, which is where the extra seconds of a brand-new install go.

Serving projects

What these numbers are not

• Not a comparison. A fair one compares whole workflows on the same machine: how much of your disk a VM takes, how long a per-project container set needs to come up, what a file-sync daemon costs in CPU. Those are other people's numbers and they are on their sites;
• Not cold starts. Pulling 600 MB of images takes as long as your connection does; a first database start also writes its data directory;
• One machine, one run. An M3 laptop with fast storage. Memory of a database grows with its workload, and docker stats memory includes page cache;
• Not a promise about your project. 26 ms is a Laravel boot plus a health route. Your application's response time is your application's business.

Reproduce it

With a running stack, from the repository:

# images
docker images --format '{{.Repository}}:{{.Tag}} {{.Size}}' | sort

# idle memory, after a minute of doing nothing
docker stats --no-stream --format '{{.Name}} {{.MemUsage}}'

# time from a stopped stack to a database that answers
laser stop
time ( laser up -d; until laser mysql -e 'select 1' >/dev/null 2>&1; do sleep 0.5; done )

# projects served, and how long the scan takes
find "$PROJECTS_DIRECTORY" -maxdepth 2 -name artisan | wc -l
time laser scan

# response time through nginx and php-fpm
curl -sS -o /dev/null -w '%{http_code} %{time_total}s\n' -H "Host: myapp.$SUFFIX" http://127.0.0.1/

If your numbers differ, they are the ones that matter: this page exists so that nobody has to take a marketing claim on faith.