DMCP study materials
DMCP study resources

DMCP: Cloud and Virtualization

DMCP · Work through your exam, one topic at a time.

All guide sections

Cloud and Virtualization

Cloud and Virtualization covers the hypervisor layer and its cloud equivalents: VM lifecycle, migration, HA and DRS admission control, resource pools, and the service and deployment models that decide what you actually operate versus what the provider operates.

Core rules

  • Live migration needs compatible CPU feature sets (EVC or a common baseline), shared or fast-migrating storage, and a network carrying the memory state. Without CPU masking, a mixed-generation cluster pins VMs to the oldest host.
  • HA restarts VMs on surviving hosts after a failure - a reboot measured in minutes, not a live migration. Admission control reserves capacity for that restart, and the policy (percentage, slot, or N+1) determines how much.
  • Resource pools: shares are relative and only matter under contention; reservations are guaranteed and can block power-on; limits cap burst. Overlapping reservations across pools are the classic reason a VM will not start despite apparent free capacity.
  • Snapshots are short-term rollback, not backup; a long snapshot chain on a thin-provisioned datastore is a capacity incident waiting to happen.
  • Service models: IaaS leaves you the guest OS and above, PaaS leaves you the application, SaaS leaves you the configuration. Migration options - rehost, replatform, refactor - differ in cost and in how much operational burden moves.

Worked example

A cluster has 6 hosts, each 2 x 16 cores (32 cores) and 512 GB RAM, CPU overcommit 4:1, and N+1 admission control. After losing the largest host, 5 x 32 = 160 cores remain, giving 640 vCPU of capacity at 4:1 against 500 vCPU of demand. Memory after failure is 5 x 512 = 2,560 GB against 2,200 GB of demand, which also fits. The reasoning to show is that admission control is evaluated post-failure, not on the healthy cluster.

Traps

  • Promising zero-downtime failover from an HA cluster.
  • Ignoring EVC and then wondering why migration fails onto newer hosts.
  • Sizing a cluster without reserving the N+1 host capacity.
  • Treating a datastore's thin-provisioned free space as real free space.