Initial problem
Cloud plans advertise vCPUs, RAM, and NVMe storage, but those labels say little about the performance a customer receives on a real shared host. CPU steal time, noisy-neighbor variance, random I/O, and fsync latency stay invisible until a production workload is already affected. Existing comparison tables repeat specification sheets instead of publishing measurements that can be reproduced or disputed.
Implementation
ProviderBench ships as a static Go binary that runs a complete benchmark in roughly ninety seconds and writes a portable JSON report. It measures compute, memory, disk, network, and virtualization signals with documented source code and copy-to-run commands. Managed Hetzner campaigns can provision multiple fresh hosts, spread samples across physical infrastructure, execute repeated measurements, and tear everything down. The Next.js website publishes dated raw results and keeps provisional datasets visibly separate from ranked comparisons.
Outcome
My contribution
Impact
What mattered
Dependency-free Go CLI with a full run in roughly ninety seconds
Measurements for CPU throughput, steal time, memory, random I/O, fsync, and network
Raw dated JSON reports that remain reproducible and independently disputable
Managed Hetzner campaigns across fresh instances and repeated samples
Public metric methodology with direct links to the benchmark source
Copy-to-run commands connecting every published number back to the CLI
Insights
The live comparison connects every metric to its method, source, and CLI command
One static Go binary runs the benchmark and produces a shareable JSON report
Technical frame
The technical frame sits at the end on purpose: it shows the engineering substance behind the solution.
Next project
An unofficial product concept for a living digital twin.
View project