We would rather run a small number of well-provisioned facilities than a map with many dots on it. Every site can carry its neighbour's load.
Packaging deliberately runs in one place. Distributing a transcode farm across four sites buys resilience we do not need and creates a consistency problem we would rather not have.
Each site takes two transit providers and connects to the local exchange. Announcements are identical everywhere; we do not do traffic engineering by selectively withdrawing prefixes, because it makes debugging someone else's routing problem impossible.
Prefixes are validated against RPKI. Invalid announcements are dropped rather than accepted with a lower preference.
Open policy at every exchange where we are present. No ratio requirement, no fee. We need a registered ASN, current registry objects and a NOC address that answers.
Requests: peering@vostokhub.com
We publish what we measure, because a hit ratio without a definition is marketing.
Hit ratio is counted per resource on segment paths only; manifests are excluded because their volume distorts the number upward. Availability is measured from probes outside our own network at thirty-second intervals, and a request counts as failed if it does not complete, not merely if it returns an error code.
Monthly reports include the raw counters, not just the derived percentages, so a customer can recompute anything we claim.