Capacity planning

The screen answers "is there room for this addition?" — the question asked today by a project manager and answered from memory by one person. It is a form, not a wizard: three groups of fields and an answer, on one screen.

What is filled in

Group	Fields
What is added	Count, vCPU each, memory each, disk each
Where	Any cluster · a specific cluster · a specific site
Physical addition	Chassis height and power each — optional, and what opens the rack answer
When	A date. Empty means today

The By an existing system selector fills the fields from the average machine of a system already running in the estate, and the sample size travels with it: "based on 12 existing machines · average 4 vCPU · 16 GB". An average taken over three machines and an average taken over eighty are not the same claim.

The cluster answer

For each cluster in the chosen scope: memory required · available after failure · remaining · utilization after · vCPU after, and the verdict. The verdict is three-valued:

Verdict	Meaning
Fits	The addition fits while redundancy is preserved
Fits without redundancy	Fits only if redundancy is ignored — a host failure after the addition leaves the cluster short
Does not fit	There is no room even without redundancy

The middle value gets a name of its own and is never folded into a green yes, because it is the answer somebody will otherwise arrive at by accident. The list is sorted by fit: those that take it with room to spare, then those that take it without redundancy, then those that do not.

When no cluster fits, the screen does not stop at "no": it names the nearest cluster and the gap in GB.

The rack answer

Racks are checked only once a chassis height is filled in under "Physical addition" — a virtual machine occupies no rack units and adds no power, so it has no physical answer.

Per rack: the free run, Space?, power after the addition, Power? and Fits?.

• A rack with a large enough run but no circuit left is marked Blocked on power. This is the half that breaks plans: a free run is not free space once the circuit is finished, and when there are such racks the screen counts them and names them.
• A rack with no electrical record gets Not checked in the power column, and its ranking falls back to least-loaded-first — stated plainly, not dressed up as headroom.
• When no rack has a large enough run, the screen names the largest run that does exist and the rack it is in, rather than saying "no space", which is usually false.
• Twelve racks are shown; the number remaining is stated.

What the screen does not do

This is a feasibility check, not an optimizer. It does not choose a cluster, does not rank by cost, does not rebalance, and it never moves a machine. The system is the change-control record; a tool that both approves change windows and moves workloads itself has removed the separation it was bought for. The output is a sentence a person acts on.

The vCPU:pCPU ratio after the addition is shown and not graded, here as everywhere else in the module.

A scenario is a URL

Nothing is saved. Every field is written into the address, and a Copy link to scenario button copies it — so a scenario is shared by pasting a link, which is how it will actually be shared. No second saving mechanism with its own permissions and sharing model was built for an object whose half-life is one meeting.

The date and the forecast

The When field is disabled until there is enough of a daily measurement series, and it says why. Accepting a date and silently ignoring it is worse than refusing it.

When it is active, the answer is computed on the projected allocation on that day, and the method is printed on the screen: ordinary least-squares linear regression over the observed series. There is no seasonality, no confidence band and no richer model — a trend line implies a model, and a reader who cannot see which model is entitled to assume the expensive one.

The forecast refuses, and that is most of its value. It will not draw a line when:

Condition	What is said instead
Fewer than 14 measurements	How many are needed and how many were collected
The series spans fewer than 21 days	How many days it spans
Fit quality below R² = 0.5	"The series does not behave like a line", with the points shown and no line
Allocation is not growing	"There is no runway to compute" — a falling series has no runway, and a flat one has an infinite one
The threshold is already crossed	It says so, rather than "0 days"
Growth too slow to cross the threshold within a decade	It says so. A runway past a decade is arithmetic, not information

All four floors — point count, span, fit quality and the series window — are editable in settings.

Permission

This screen has a permission of its own, separate from the view permission for the rest of the module: planning is a capability beside viewing, not a rung on a ladder. Anyone holding it also sees the other capacity screens.