Five sites producing 217 MB of backups a night. A mirror uploads all of it; a deduplicating tool uploaded 4.4 MB. Plus the deletion trap in rclone sync and what your bucket keys actually expose.
A Redis object cache took a WordPress page from 64 SQL queries to 9 and made it 2.5 ms slower. Redis sessions looked 2.7x faster and lost 13 of 20 concurrent writes.
Deployed on a live server already running five production sites, scaled to 60 monitors, and measured: 104.7 MiB idle, 359 bytes per heartbeat, 30 MB of disk a day, and a status page serving 878 requests per second.
Seven applications benchmarked on one 2 vCPU VPS. Six land between 11 and 43 requests per second per core regardless of language or framework, and caching is the only lever with an order of magnitude in it. With the isolation model, the economics with your own hours priced in, and the application-layer incident that mattered more than the perimeter.
Hardening steps ordered by measured effect, from a live server that logged 1,422 failed SSH attempts in one week. Includes the two silent fail2ban failures we found on our own machine, and the sysctl audit showing nine of ten hardening settings are already Ubuntu defaults.
Measured on a live n8n 2.37.10 deployment limited to 1 vCPU: 11.4 executions per second, a 402 MB idle footprint, 16.9 second cold start, and 3,563 bytes of database growth per execution. Complete Compose file, nginx reverse proxy, queue mode, and five documented failure modes.
Measured on a live 2 vCPU KVM VPS running Laravel 13.30.1: 16.4 req/s without OPcache versus 190.8 req/s fully tuned. Complete nginx, PHP-FPM and systemd configuration, an atomic zero-downtime deploy script, and five documented failure modes.
Measured on a live 2 vCPU KVM VPS: 19.9 req/s uncached versus 749.6 req/s with an nginx FastCGI microcache. Complete nginx and PHP-FPM configuration, a marketplace script security audit protocol, and four documented failure modes.
↑↓ navigate ↵ open esc close