3.8.2. Resource Allocation and Cloud Models
💡 First Principle: VM resources (CPU, memory, disk, NIC) are allocated from the host's physical resources. The hypervisor multiplexes these resources among all running VMs. Overprovisioning—allocating more virtual resources than physical resources exist—is possible and sometimes acceptable, but requires careful monitoring to avoid contention.
Resource Allocation
| Resource | Allocation Method | Overprovisioning Risk |
|---|---|---|
| CPU | vCPUs assigned to VMs; host schedules access to physical cores | CPU contention causes latency for all VMs; avoid >4:1 vCPU:pCPU ratio for production |
| Memory | Reserved or shared; hypervisor may balloon or swap VM memory | Memory overprovisioning causes VM swapping (severe performance impact) |
| Disk | Thin provisioned (grows on demand) or thick (preallocated) | Thin provisioning risks datastore exhaustion if all VMs grow simultaneously |
| NIC | Bandwidth shared among all VMs on same physical NIC | Network-heavy VMs compete for bandwidth; use multiple physical NICs |
Scalability
Virtualization enables rapid scaling: a new VM can be deployed in minutes from a template. Horizontal scaling (adding more VMs) is often faster than vertical scaling (adding more resources to existing VMs) in virtual environments.
Cloud Models
| Model | Who Owns Hardware | Who Manages | Example |
|---|---|---|---|
| Public cloud | Cloud provider | Shared responsibility | AWS, Azure, Google Cloud |
| Private cloud | Your organization | Your IT team | On-premises VMware vSphere |
| Hybrid cloud | Both | Both | On-prem for sensitive data; public for burst workloads |
Management Interfaces for VMs
- VMware: vCenter Server (centralized management), vSphere Client (web UI), ESXi Shell, PowerCLI
- Microsoft Hyper-V: Hyper-V Manager, Windows Admin Center, PowerShell
- All platforms: REST APIs for programmatic management and automation
⚠️ Exam Trap: Thin provisioning is not the same as overprovisioning deliberately. Thin provisioning is a storage efficiency feature; overprovisioning is a capacity planning decision. Thin-provisioned VMs that grow beyond the datastore's physical capacity cause all VMs on that datastore to crash when the datastore fills.
Reflection Question: A host with 256GB RAM runs 30 VMs, each configured with 16GB RAM (480GB total allocation). The hypervisor reports no performance issues currently. Is this overprovisioning a problem? When would it become a problem?
