SQL Mechanics

Your SQL Server. Tuned to Perfection.

Success Story: High-Performance SQL Server on Azure Using Ultra Disks and Constrained vCPU VMs

—

by

sqlmechanic
in

Overcoming Azure’s SQL Server Storage Performance Limitations:

For years, running SQL Server on Azure VMs came with a major challenge: slow disk performance. Traditional Azure managed disks often forced SQL workloads to compromise between capacity and speed. Even the largest 32 TB Premium disks delivered insufficient throughput compared to modern on-premises SSD solutions.

Most organizations were forced to put both DATA and LOG files onto a single 8 TB Premium SSD—the only disk option with some performance. Attempting to separate LOG files onto smaller disks resulted in IOPS bottlenecks and inconsistent write performance.

This long-standing storage limitation prevented many businesses from confidently migrating SQL Server to Azure.

Azure Ultra Disks: A Breakthrough for SQL Server Performance

The arrival of Azure Ultra Disks changed everything.

For the first time, Azure storage allowed fully independent configuration of:

  • Disk Size
  • IOPS
  • Throughput

This made it possible to deploy, for example, a 100 GB disk with 30,000 IOPS and 1 GB/s throughput—ideal for high-performance SQL Server LOG volumes.

With Ultra Disks, Azure became a truly viable platform for enterprise-grade SQL Server performance.

Client Objective: Migrate From Private Hosting to Azure

A long-term client wanted to move from privately hosted hardware to Azure to:

  • Lower infrastructure and staffing costs
  • Improve reliability
  • Gain cloud flexibility
  • Maintain or exceed existing SQL Server performance

The primary concern: Would Azure deliver the low-latency storage required for their SQL Server workload?

With Ultra Disks and recent VM generations, the answer became yes.

Our Architecture: Optimized for Performance & Licensing Efficiency

To achieve maximum SQL performance while minimizing SQL Server licensing costs, we designed a solution using an AMD EPYC (Milan)–based Constrained vCPU VM:

  • 32 physical cores, constrained to 8 vCPUs → significant SQL Server Enterprise license savings
  • 192 GB RAM for buffer pool optimization

Storage Configuration

  • DATA files: 8 TB Premium Managed Disk
  • LOG files: 1 TB Ultra Disk
    • 20,000 IOPS
    • 1 GB/s throughput
  • tempdb: High-speed local ephemeral SSD

This layout ensures the highest possible write performance for LOG files while maintaining predictable DATA throughput.

Results: Sub-Millisecond Latency and Long-Term Stability

Post-migration performance exceeded expectations:

  • 0.25 ms average write latency on LOG files
  • Consistent throughput under peak transactional load
  • No storage bottlenecks
  • Smooth, predictable performance even during heavy reporting periods

For high availability, the client implemented SQL Server Failover Cluster Instances (FCI)—a cost-efficient choice requiring only one set of shared storage.

Multi-Year Reliability

The system has now been running for several years with:

  • No storage-related performance degradation
  • No unplanned downtime
  • Stable, predictable SQL Server behavior

Why This Matters for Azure SQL Server Migrations

This success story demonstrates that:

  • Ultra Disks eliminate historical Azure SQL storage limitations
  • Constrained vCPU VMs dramatically reduce SQL licensing costs
  • Azure can match or exceed on-prem SQL Server performance when properly architected

If your organization is evaluating a move to Azure, modern storage options now make high-performance SQL Server deployments not just possible—but highly cost-effective.