SQL Server Workloads in Azure with Everpure Cloud Dedicated Design Validation

Microsoft Platform Guide

Audience
Public
Source Type
Documentation

This design validation evaluates the performance benefits of deploying SQL Server workloads on Azure Virtual Machines using Everpure Cloud Dedicated (V20 with Premium disks) compared to Azure Premium SSD v2 Managed Disks. The testing methodology simulates realistic transactional and analytical database workloads to characterize storage performance and throughput.

HammerDB Benchmark Suite

HammerDB is an industry-standard, open-source database benchmarking tool designed to simulate realistic database workloads, measuring the performance of transactional (OLTP) and analytical (OLAP) databases. It provides standardized benchmarks such as TPC-C and TPC-H, widely recognized for evaluating database system performance and scalability.

Workloads and Methodology

This design validation utilizes the following common environment components and software:

Component

Detail

Azure Virtual Machine

Specification: Standard E48as v5 (48 vCPUs, 384 GiB RAM, 71,000 IOPS limit)

Operating system: Windows Server 2025 Datacenter

Networking: Private IP with accelerated networking enabled

Azure managed disks:

  • TempDB – Ultra disk LRS – Max IOPs=3000, Max throughput (MBps)=200

  • SQL_data01 – Premium SSDv2 LRS – Max IOPs=20000, Max throughput (MBps)=250

  • SQL_data02– Premium SSDv2 LRS – Max IOPs=20000, Max throughput (MBps)=250

  • SQL_data03– Premium SSDv2 LRS – Max IOPs=20000, Max throughput (MBps)=250

  • SQL_data04– Premium SSDv2 LRS – Max IOPs=20000, Max throughput (MBps)=250

  • SQL_log– Premium SSDv2 LRS – Max IOPs=20000, Max throughput (MBps)=250

Pure Cloud Block Store

V20 with Premium SSD (CBS-V20MP2R2) with Purity 6.8.5.

Volumes provisioned:

  • CBS_SQL_data01

  • CBS_SQL_data02

  • CBS_SQL_data03

  • CBS_SQL_data04

  • CBS_SQL_log

iSCSI sessions: 4 (2 per controller)

Multiple connected sessions policy: Least Queue Depth (LQD)

Microsoft SQL Server

SQL Server 2022 Developer Edition

TPCROC-C database: 4 data files (128GB each) and 1 log file, each within its own volume.

TPCROC-H database: 4 data files (256GB each) and 1 log file, each within its own volume.

Max Degree of parallelism: 8 Cost threshold for parallelism: 50

Memory limits (TPROC-C): 64GB

Memory limits (TPROC-H): 100GB

HammerDB

Version 4.12

Note: Actual performance outcomes may vary depending on database structure, workload characteristics, and Azure infrastructure specifics. It's advisable for organizations to conduct environment-specific testing to validate these benefits.
Transaction Processing Workload (TPROC-C)

This workload simulated a high-intensity transaction processing environment typical for OLTP databases. The comparative metric used was new orders per minute (NOPM). The schema was built and then backed up, being restored to the relevant storage type being evaluated.

The following configuration was used for the transaction processing workload:

  • Number of warehouses: 5000

  • Virtual users: 200

  • User delay (ms): 500

  • Repeat delay (ms): 500

  • Iterations: 1

  • TPROC-C driver script: Timed Driver Script

  • Total transactions per user: 10000000

  • Checkpoint when complete

  • Minutes of ramp-up time: 5

  • Minutes for test duration: 30

  • Warehouses: Use all warehouses

  • Profile: Time profile

Results and Analysis

In the TPC-C transactional processing benchmark, Pure Everpure Cloud Dedicated achieved a transactional throughput of 223,214 new orders per minute (NOPM), outperforming Azure Premium SSD v2 managed disks, which reached 159,732 NOPM. This represents approximately 40% performance improvement using Everpure Cloud Dedicated.

Notably, the Azure managed disks' performance was limited by the Azure VM's maximum IOPS constraint (71,000 IOPS), indicating that Everpure Cloud Dedicated's performance scalability extends beyond the inherent limitations of the VM's disk infrastructure. These results highlight the ability of Everpure Cloud Dedicated to independently scale storage performance, providing significant transactional performance benefits without necessitating more powerful or costly virtual machine configurations.

Figure 1. TPROC-C performance results comparing Everpure Cloud Dedicated and managed disk
Decision Support Workload (TPROC-H)

This workload simulated a complex analytical and decision-support environment, representative of OLAP workloads common in data warehousing scenarios. The comparative metric used was queries per minute (QPM). The database schema was generated and loaded at a defined scale factor, backed up, and subsequently restored onto the evaluated storage solutions before executing the benchmark queries.

The following configuration was used for the analytical processing workload:

  • Scale factor: 1000

  • MAXDOP: 8

  • Build option: Clustered ColumnStore

  • Virtual users: 1

  • Total query sets per user: 1

  • Iterations: One full run of the query set

  • Execution profile: Sequential execution of queries

In the TPC-H analytical processing benchmark, Pure Everpure Cloud Dedicated achieved a throughput of 1.45 QPM, significantly surpassing Azure Premium SSD v2 managed disks, which delivered 0.75 QPM. This result represents approximately a 93% improvement in analytical query performance when using Everpure Cloud Dedicated.

The results demonstrate Everpure Cloud Dedicated's capability to handle intensive analytical workloads. Organizations leveraging Everpure Cloud Dedicated for SQL Server analytical workloads can therefore expect substantially improved query responsiveness and reduced processing times compared to similar managed-disk configurations.

Figure 2. TPROC-H performance results comparing Everpure Cloud Dedicated and managed disk