Test identity
Disposable production-like KVM guests are created with the same CPU/RAM/disk envelope as the published VPS 2 and VPS 4 plans. The guest OS, kernel, resource limits, host class and run timestamp are recorded in the raw artifact.
Reproducible evidence
VPS benchmarks with the settings attached.
A benchmark number without the plan, guest, node context, command, duration and timestamp is marketing decoration. We publish the method and raw archive together.
CPU
sysbench CPU runs use a fixed max-prime and duration recorded with each result. vCPU is shared/scheduled capacity, so runs can vary; we preserve repeated results instead of selecting only the fastest sample.
Storage
fio jobs use explicit block size, queue depth, I/O engine, read/write mix, direct-I/O setting, file size and duration. The test file stays inside the disposable guest disk. We do not call a short synthetic fio run an application-performance guarantee.
Network and latency
Network tests record the endpoint and protocol. Canadian latency observations name the actual source location we control; they are not generalized into a Canada-wide latency promise.
Archive everything
Raw results live under data/benchmarks/vps/. Failed, noisy and unfavourable runs are retained with their status so later comparisons can see variance and methodology changes.
Latest certified run
VPS 2 and VPS 4 evidence.
Run ID: gate15-2026-09-01-bhs · observed 2026-09-01
complete
Single-thread CPU 1,599.55 events/s; all-vCPU CPU 1,599.55 events/s; direct sequential write 2.28 GB/s; direct sequential read 6.93 GB/s; 4K 70/30 mixed read 79,728 read IOPS; same-region BHS download median 295.1 MB/s.
complete
Single-thread CPU 1,579.42 events/s; all-vCPU CPU 3,169.40 events/s; direct sequential write 2.55 GB/s; direct sequential read 7.10 GB/s; 4K 70/30 mixed read 86,935 read IOPS; same-region BHS download median 285.8 MB/s.
Evidence note
Results are short synthetic observations, not an uptime or workload guarantee.
Evidence note
Network numbers are same-provider/same-region BHS guest-to-public-storefront observations through the QA guest NAT path, not Canada-wide Internet throughput.
Evidence note
The first VPS 2 measurement transport completed but its stdout was not retained; no claim is derived from that unretained attempt. The retained raw run below is the published VPS 2 evidence.