PERFORMANCE & ENGINEERING
A small stack.
Measured in operation.
A custom web and mail stack written in V, with replaceable TLS implementations and explicit resource limits.
Snapshot recorded 2026-10-21 18:19:18 UTC. Live freshness requires JavaScript; check the timestamp before interpreting these observations.
Snapshot generated 2026-10-21 18:19:18 UTC.
Observed from this server
Service checks originate on the VPS. Outbound tests measure connectivity to selected destinations. These measurements do not establish website reachability from visitors’ networks.
- website: 18 successful / 18 samples
- outbound-dns: 18 successful / 18 samples
- outbound-https: 18 successful / 18 samples
- mail-reception: 18 successful / 18 samples
Daily observations
Bars show successful and failed check samples. Empty dates are gaps, not successful observations.
Resource use
Memory is measured per service control group on Linux, including charged cache and child processes. The web service is part of the hosting total. Short-lived jobs may finish between samples. Values are not process RSS.
Server processing
Server processing measurements are unavailable or have too few samples.
Processing durations cover HTTP request handling before socket writes, excluding TLS and client transfer. Histogram bounds are reported since the web process started. The final bucket means at least 1.6384 seconds. Idle periods can make this measurement stale.
How timings are measured
Median and 95th-percentile durations cover successful synthetic checks in the last 24 hours, including bounded worker startup. At least 20 successful samples are required. Failures are counted separately. HTTP checks stop after the expected content is found; durations are neither full browser load times nor server processing times.
Checks normally run every five minutes. Timestamps use UTC; durations use a monotonic clock. Missing intervals and clock changes are reported explicitly. No visitor identifiers or individual visitor request records are collected for this page.
Benchmarks
Isolated benchmark results are available as a complete JSON report when published.
The comparison serves an identical small file over TLS 1.3, with fresh connections, no compression, three repeats and concurrency 1, 8 and 32. Client and server share the workstation. Results describe this workload only; memory uses proportional process-tree accounting, distinct from the live cgroup measurements above.
Engineering choices
- V implementation with operating-system adapters.
- Replaceable TLS backends with certificate verification.
- Separated service permissions and bounded network operations.
- Encrypted recovery backups and a tested restore procedure.
- A fixed public export schema; diagnostic logs and mail information remain private.
These are design and operational properties, not proof that custom software is more secure. Platform support and security claims must follow actual validation.