Azure VMs and scaling
VM basics
IaaS compute: full OS control, no physical hardware. Size means vCPU
- RAM + network throughput; image is the OS template, Microsoft’s or your own.
Billing runs while the VM runs. A deallocated VM stops compute charges (disks still bill).
The SLA ladder (memorize)
- Single VM with all Premium SSD or Ultra disks: 99.9%
- Availability Set (fault + update domains): 99.95%
- Across Availability Zones: 99.99%
An Availability Set is redundancy inside one datacenter: fault domains guard against hardware failures, update domains against planned maintenance. Manual redundancy, no scaling.
Availability Zones are separate datacenters: the top VM uptime tier.
Scaling
Vertical (up and down) resizes the VM and requires downtime. Horizontal (out and in) adds or removes instances with zero downtime: the modern default.
VM Scale Sets (VMSS) are a group of identical load-balanced VMs that autoscale on CPU, traffic, or custom rules. VMSS is automated scaling; availability sets are redundancy only.
Azure Virtual Desktop serves managed Windows desktops and apps from the cloud for remote workers, contractors, BYOD, and call centers; pooling shares VMs to cut cost.
VM series
- B, “Budget”: burstable, dev and test.
- D, “Daily use”: general purpose.
- E, “Elephant memory”: memory-optimized (SAP, in-memory analytics).
- F, “Fast CPU”: compute-optimized (batch, game servers).
- M, “Massive memory”: up to 12TB RAM, SAP HANA.
Disks and cost
The disk ladder: Standard HDD (cheap), Standard SSD (general production), Premium SSD (high IO, and required, along with Ultra, for the 99.9% single-VM SLA).
Spot VMs run up to 90% off with a 30-second eviction notice. Reserved Instances save up to 60% (1 year) or 72% (3 years) for always-on workloads.
Azure Advisor flags oversized or idle VMs for free.