Skip to main content

Resources and Capacity

Infrastructure > Resources is a capacity cockpit: where the estate stands, what needs a decision, and how much room is left, per cluster, per node and per storage pool, with a forecast drawn from up to a year of history. It is an Enterprise feature, included with the Green IT metrics.

Pick All clusters or one cluster at the top left. The PDF icon at the top right downloads a capacity report built from the same figures, and the refresh icon reloads them.

Resources cockpit for all clusters: an infrastructure health score of 90 out of 100 rated Excellent, with active VMs, efficiency and alerts, the CPU, RAM and storage usage bars, the health score trend over six daily snapshots, the To handle bar with 12 items including 5 critical, and the Capacity, Simulation, Efficiency, Sustainability and AI recommendations tabs
The health score and its history, what needs a decision, then five tabs.

Health score​

The score out of 100 sums up usage, efficiency and active alerts for the selected scope. Its trend card keeps one snapshot per day and per tenant, with the lowest and highest values of the period. The score and the To handle list use different rules: the score can read Excellent while the list holds critical items, so read both.

To handle​

To handle lists what needs a decision, most urgent first. It is folded by default; its badge counts the items and the critical ones.

To handle unfolded: storage pools over their threshold or full in a few days at their current growth, and clusters whose RAM is unevenly spread across nodes, each tagged Capacity
Each item opens the tab that explains it, scrolls to the section and highlights the row.

It raises, among others:

  • a cluster that would not survive the loss of its largest node (N+1);
  • clusters and storage pools over their threshold, or that reach it soon at their current growth;
  • offline nodes;
  • RAM unevenly spread across the nodes of a cluster;
  • allocation held by guests that have been idle for a long time.

Thresholds come from Settings > Alert thresholds.

Capacity​

Capacity tab: the Evolution and Projections chart of CPU, RAM and storage over 61 days with its projection, the 90d, 180d and 365d history windows, and the 30-day forecast with three dials, CPU 6 percent beyond a year, memory 55 percent 58 days before 90 percent, storage 56 percent 95 days before 90 percent, with the measured error
The usage history with its projection, and the time left before each resource reaches the critical threshold.

Evolution & Projections draws CPU, RAM and storage usage and extends them as dashed lines. The 90d, 180d and 365d buttons pick how much history feeds the forecast; switching does not query Proxmox again, and the choice is remembered per browser. A window longer than the available history is greyed out, with the number of days available in its tooltip, and the chart header shows the range it actually uses. On Proxmox VE 9, the yearly history holds about 360 days.

The 30-day forecast shows, for each resource, today's usage, its value in 30 days and the time left before the critical threshold, up to a year ahead. The forecast weights the last 30 days more than older ones, starts from the median of the last three days so that a one-day spike does not move it, and follows the weekly pattern. Its measured error is obtained by rerunning the forecast from past days and comparing it with what really happened.

Below the charts, the tab breaks capacity down per cluster (usage, vCPU:pCPU ratio, N+1, imbalance and runway), per node (load and allocation) and per storage pool (usage and growth). A shared pool is counted once per cluster, a local pool once per node.

The AI forecast reading puts the figures into words. It uses the provider set in AI Insights, and falls back to a rule-based reading without one. The model only ranks and words what the platform computed: every figure comes from the calculation, and it cannot rate a risk lower than the rules do.

Simulation​

The what-if simulator tests a growth or a failure on one cluster before it happens.

What-if simulator on PVE-DR: a VM template of 4 vCPU and 8 GB, an accepted vCPU:pCPU ratio of 4, a scenario adding 10 VMs; today fits with RAM at 48 percent, after the scenario RAM reaches 162 percent and is over threshold, with no VM left that still fits
Add VMs or nodes, or lose the largest node, and compare today with the result.

Choose a VM size, the vCPU:pCPU ratio you accept, then the scenario: VMs to add, nodes to add, and Lose the largest node. New VMs count at their full RAM and at the CPU today's guests draw per vCPU; added nodes copy the largest online node.

Efficiency​

Overprovisioning analysis: a CPU vCPU to pCPU ratio of 1.9 to 1 with 89 vCPUs allocated on 48 cores, a RAM ratio of 0.60 to 1 with 85 GB allocated on 141 GB, and the optimization potential of 86 recoverable vCPUs, 4 GB of recoverable RAM and 10 VMs to rightsize
Allocation against physical capacity, and what rightsizing would recover.

The tab compares what guests are allocated with the physical capacity and with what they really use, lists the VMs to rightsize and the top consumers, and shows the network throughput.

Sustainability and AI recommendations​

Sustainability shows the energy, carbon and cost figures described in Datacenters and Green IT.

AI recommendations builds an optimization plan from the same figures, with the provider set in AI Insights. The model is called only when the tab is opened, its reading is kept 15 minutes per scope and language, and items the model leaves out are added back.