VM Placement & Live Migration

Fleet Controls

VM arrival rate 12/min

Physical hosts 6

Consolidation threshold 20%

Placement strategy

Fleet Status

Hosts powered

0 / 0

Fleet utilization

Live migrations

VMs running / rejected

0 / 0

1-core VM 2-core 3-core 4-core

How it works

Each physical host has a fixed capacity of 16 vCPU units. Every VM that arrives asks for a random 1–4 units. A placement scheduler chooses which powered-on host gets it:

Consolidate (First-Fit):
  place on the first host with
  free(h) >= size(vm)

Spread (Best-Fit-balance):
  place on the host with the
  largest free(h) >= size(vm)

Utilization of a host is U(h) = used(h) / capacity(h). Every few seconds a consolidation pass checks each powered host: if U(h) is below the threshold slider and every VM on it can be re-homed onto other powered hosts (enough free capacity exists elsewhere), those VMs are live-migrated — moved to a new host while the simulation keeps running — and the now-empty host is powered down to save energy. If demand later exceeds what powered hosts can hold, a sleeping host wakes back up.

Consolidate packs VMs tightly onto as few hosts as possible, so idle hosts sleep sooner — good for power efficiency, worse for headroom against sudden bursts.
Spread balances load evenly across every powered host — steadier per-host headroom, but hosts rarely go idle enough to sleep.
A VM that finds no host with enough free capacity, and no sleeping host to wake, is rejected — this is exactly what capacity planning in a real cloud provider tries to avoid.

This is a simplified model of what a hypervisor scheduler (VMware DRS, OpenStack Nova, KVM live migration) does continuously in a real data center.

Host platform High utilization Idle / sleeping

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.

VM Placement & Live Migration in a Cloud Fleet