What
Add extra streaming capacity for mercator-talks on flat heads by renting cloud GPU machines in or near Belgium and joining them to the fleet as bodies in a cloud district.
Why
Venue bodies are a fixed pool. #504 multi-stream slots raise capacity per body, but the number of physical GPUs does not change. Cloud GPU bodies give elastic capacity for peaks and for venues that have heads but no local body.
Requirements
- Cloud bodies run the standard body stack (Windows, Sunshine, hydrabody, WireGuard mesh to the HydraGuard hub) or a documented variant, in their own cloud district.
- Region and provider selection is driven by measured latency and jitter from the actual venues, not by datacenter location claims.
- Part of this issue: a repeatable procedure to test actual latency (RTT, jitter p50/p95/p99, loss) from each hydra venue to each candidate cloud region, before any commitment to a provider.
- The test procedure must produce the same metrics as #397/#402 (network-quality telemetry: jitter, loss, reordering, RTT, percentiles) so venue qualification and cloud region qualification use one method. Reuse or extend #42 (latency proximity checker, done) where it fits.
- mercator-talks concurrency constraints from #504/#511 (per-slot audio and mic device selection) apply to cloud bodies the same as to venue bodies. #513 hydramark gives the stream_capacity benchmark per machine type.
Deliverable
A reviewed plan on this issue (plan field), then implementation.
Related: #42, #397, #402, #504, #511, #513.
Repo
github.com/cederikdotcom/hydrahyperscaler (v0.1.0) is the program repo: PLAN.md (kept in sync with this plan field), runbooks (smoke-test-grid, venue-latency-campaign, cloud-body-enrol-retire), testbooks, and the campaign results under docs/testbooks/region-qualification/results/.