Capacity measurement and consolidation
A consolidation ratio is the result of measurement, not a target. CPU, memory, disk latency and network use are collected as time series; the gap between peak and average sets the overcommit limit.
- Measurement covers at least one full business cycle
- Peaks are reported separately from averages
- Memory runs out before CPU in most estates
- Disk latency sets the practical consolidation ceiling
How licensing shapes architecture
Virtualization decisions change licensing cost directly. For software licensed per core, cluster design, placement rules and node size determine the bill. Design the architecture first and the licensing second, and the invoice becomes a surprise.
- Placement rules keep licensing boundaries intact
- Node size is chosen together with core counting
- Audit readiness is maintained continuously
- Refresh is aligned with the contract term
High availability and maintenance
A cluster must absorb the loss of a node without business impact. Redundancy headroom, maintenance windows and live migration capability are planned together, and the patch calendar is built against that headroom.
- Headroom is calculated against the busiest hour
- The patch calendar is tied to cluster capacity
- Live migration separates maintenance from downtime
- The storage path carries no single point of failure
Refresh and exit planning
Hardware and platform versions do not age on the same calendar. The refresh plan considers end-of-support dates alongside budget periods, and the cost of exit is known up front in case the platform changes.
- End-of-support dates tracked on one calendar
- Refresh waves aligned to budget periods
- Exit cost measured alongside the cost of staying
- Data portability treated as a design criterion