Business outcomes
Every number below is measured off this build, cited to docs/phase-log.md, not a
projection or a vendor benchmark. The design rewards spinning hosts up and down rather
than holding static capacity.
Elasticity in minutes, not procurement cycles
| Measurement | Result |
|---|---|
OpenShift IPI install, end to end |
44m35s |
Bare-metal |
~20 min |
|
~5 min |
First test VM, create to |
18 seconds |
Windows compute-VM clone-to- |
~2-3 minutes |
All figures: docs/phase-log.md, 2026-07-09 "Phase 1 EXIT CRITERIA MET" and 2026-07-10
entries. What used to be a hardware procurement cycle is now a golden-image clone.
Burst economics favor high VM churn
The metal worker pool is the fixed cost — baseline ~$5.2/hr for this build’s cluster,
including one m5zn.metal node. Scaling that MachineSet to 0 saves ~$4/hr and
re-provisions in ~20 min. Windows compute VMs on top of already-paid metal are
near-zero marginal infrastructure cost, so the design rewards spinning hosts up and down
rather than holding static capacity for a peak that is rarely sustained — the opposite
of the static-capacity problem this replaces (see The
problem).
One platform team for containers and VMs
GitOps reconciles both estates through the same Argo CD app-of-apps model already used for containerized workloads elsewhere in the org. See GitOps for the app-of-apps layout.
Exit from hypervisor licensing volatility
No dependency on VMware’s post-Broadcom subscription-only pricing — see the vSphere row in The solution's comparison matrix. This design’s cost floor is a metal instance class (~$4/hr+), not a renewal negotiation.
Auditable supportability
One documented custom component instead of a fork or a repackaging of Symphony itself — the claim the whole design is built to defend. See the supportability spine in The solution.
Continue to Who uses what.