9.3 Azure VM High Availability: Availability Sets, Zones & Proximity Placement Groups

Azure VM High Availability: Availability Sets, Zones & Proximity Placement Groups

Designing resilient enterprise infrastructure in Microsoft Azure requires engineering against both planned platform maintenance (such as hypervisor OS patching) and unplanned hardware failures (such as physical power supply outages, top-of-rack switch failures, or datacenter disasters). Windows Server workloads deployed in Azure IaaS must be architected across availability constructs to meet strict enterprise Service Level Agreements (SLAs). This section covers the three placement constructs — Availability Sets, Availability Zones, and Proximity Placement Groups (PPGs) — and the SLA each one earns. Virtual Machine Scale Sets (VMSS), autoscale rules, and Azure Dedicated Hosts are covered in section 9.4.

1. High Availability Architecture: Availability Sets vs Availability Zones

Azure provides hierarchical availability tiers depending on whether protection is needed against rack-level hardware failure, hypervisor updates, or entire datacenter facility disruptions.

+-----------------------------------------------------------------------------+
|                  AZURE HIGH AVAILABILITY ARCHITECTURE HIERARCHY             |
|                                                                             |
|   [AVAILABILITY SET]                 [AVAILABILITY ZONES]                   |
|   - Scope: Single Datacenter         - Scope: Entire Azure Region           |
|   - Protects: Rack / Power / Host    - Protects: Facility / Flooding / Grid |
|   - SLA: 99.95% (2+ VMs)             - SLA: 99.99% (2+ VMs across Zones)    |
|                                                                             |
|   +-------------------------------+  +-----------------------------------+  |
|   | Datacenter A                  |  | Region (e.g., East US)            |  |
|   | +-----------+   +-----------+ |  | +-----------+ +-------+ +-------+ |  |
|   | | Rack 1    |   | Rack 2    | |  | | Zone 1    | | Zone 2| | Zone 3| |  |
|   | | FD 0 /UD 0|   | FD 1 /UD 1| |  | | (DC 1)    | | (DC 2)| | (DC 3)| |  |
|   | | [VM-APP01]|   | [VM-APP02]| |  | | [VM-DB01] | |[VM-DB2]| |[VM-DB3]| |  |
|   | +-----------+   +-----------+ |  | +-----------+ +-------+ +-------+ |  |
|   +-------------------------------+  +-----------------------------------+  |
+-----------------------------------------------------------------------------+

Availability Sets Mechanics:

An Availability Set is a logical grouping of VMs within a single Azure datacenter that isolates VM instances across distinct physical hardware infrastructure:

1. Fault Domains (FDs):

   • Definition: Represents a physical rack of servers sharing a common power supply unit (PSU) and top-of-rack (TOR) network switch.
   • Allocation: Azure regions provide 2 or 3 Fault Domains per Availability Set (FD 0, FD 1, FD 2).
   • Purpose: Protects against physical hardware failures, cable faults, and power interruptions. VMs in distinct FDs will never fail simultaneously due to a single physical hardware failure.
   • Managed Disk Alignment: When using Managed Disks, Azure automatically aligns the storage cluster fault domains with the compute fault domains, preventing shared storage hardware failures.

2. Update Domains (UDs):

   • Definition: A logical grouping of underlying physical hypervisors that can undergo scheduled platform maintenance or reboot simultaneously.
   • Allocation: Configurable from 1 to 20 Update Domains (default is 5).
   • Purpose: When Microsoft patches the underlying host hypervisors, only one Update Domain is rebooted at a time. The platform pauses for a 30-minute recovery window before proceeding to the next UD, ensuring remaining VM instances handle active client traffic.

3. SLA: Deploying two or more VMs across distinct FDs and UDs in an Availability Set guarantees a 99.95% uptime SLA.

Availability Zones Mechanics:

Availability Zones are physically separate datacenter facilities within an Azure region, each equipped with independent power, cooling, and high-speed fiber networking.

• Regional Isolation: Protects workloads against catastrophic datacenter-level disasters (e.g., regional power grid failure, major fire, flooding).
• Latency Profile: Inter-zone fiber networks deliver round-trip latency under 2 milliseconds, supporting synchronous database replication (such as SQL Server Always On Availability Groups with synchronous commit).
• SLA: Distributing two or more VMs across two or more Availability Zones in the same region provides Azure's highest single-region compute SLA: 99.99% uptime.
• Zonal vs. Zone-Redundant Resources:
  • Zonal Resource: A VM pinned explicitly to a single zone (e.g., Zone 1).
  • Zone-Redundant Resource: An Azure resource (e.g., Standard Load Balancer, Zone-Redundant Gateway, ZRS Managed Disk) whose underlying infrastructure spans all three zones automatically.

2. Proximity Placement Groups (PPGs)

While Availability Zones disperse VMs across different datacenters for disaster isolation, certain tightly coupled multi-tier Windows workloads require the absolute lowest possible physical network latency.

+-----------------------------------------------------------------------------+
|                   PROXIMITY PLACEMENT GROUP (PPG) TOPOLOGY                  |
|                                                                             |
|   [SAME AZURE DATACENTER / COMPUTE SPINE]                                   |
|   +---------------------------------------------------------------------+   |
|   |  Proximity Placement Group: PPG-Production                          |   |
|   |                                                                     |   |
|   |  +---------------------------+     +-----------------------------+  |   |
|   |  | Web/App Tier (IIS)        |     | Clustered Backend (SQL AG)  |  |   |
|   |  | [VM-WEB01]   [VM-WEB02]   |     | [VM-SQL01]     [VM-SQL02]   |  |   |
|   |  +---------------------------+     +-----------------------------+  |   |
|   |                \                         /                          |   |
|   |                 ===[Sub-Millisecond RTT]==                          |   |
|   +---------------------------------------------------------------------+   |
+-----------------------------------------------------------------------------+

• Architecture: A Proximity Placement Group (PPG) is a logical grouping constraint that forces Azure to provision all assigned compute resources (standalone VMs, Availability Sets, or VMSS) within the same physical datacenter cluster and network spine.
• Latency Objective: Reduces network round-trip time (RTT) between application tiers (e.g., IIS web servers calling a backend SQL Server clustered instance) to sub-millisecond levels.
• Architectural Trade-offs & Allocation Constraints:
  • Because PPGs pin VMs to a specific physical hardware cluster, future VM scaling or resizing operations may fail if the specific physical cluster runs out of capacity (OverconstrainedAllocationRequest).
  • Best Practice: When deploying a PPG, deploy all required VMs and scale sets simultaneously during initial rollout rather than incrementally over time.