About our testing approach

We compared the performance of the following cloud-based MySQL services, testing one instance each:

• Standard_Eds32_v5 Business Critical Azure Database for MySQL – Flexible Server
• Memory-optimized db.r6i.8xlarge Amazon Aurora MySQL on AWS^™
• Memory-optimized db.r6i.8xlarge Amazon RDS for MySQL, also on AWS
• db-perf-optimized-N-32 Google Cloud SQL for MySQL

Each instance had 32-core vCPUs and 256 GB of memory. We maximized disk performance, in IOPS, for each service. Azure automatically adjusted performance via the Auto-IOPS feature, and AWS used a similar function, called IO-Optimized, to maximize performance for the Aurora service. For the AWS RDS and Google Cloud services, we used larger-capacity data volumes to increase the provisioned IOPS to the instance maximums. The nominal maximum IOPS for each of the backing instances were as follows:

• Azure Standard_E32ds_v5 - 38,000 IOPS
• AWS db.r6i.8xlarge - 40,000 IOPS
• Google Cloud db-perf-optimized-N-32 - 60,000 IOPS

Using the benchmark tool sysbench, we ran an online transaction processing (OLTP) MySQL workload of 70 percent reads and 30 percent writes on each CSP service at 128 and 256 threads. We ran both workloads with 5- and 10-minute runtimes, generating four sets of results for each service.

Comparing features across services

Azure Database for MySQL – Flexible Server supports a variety of features and services, and we compared this support across all four services. Details of our comparison appear in Table 1.

*Note: While it is technically possible to create geo-redundant backups with AWS using AWS Backup,⁹ Amazon does not advertise it as a feature of Aurora or RDS, and the feature requires users to perform additional setup using an additional service.

Results of our testing

Table 2 shows the TPS and latency outputs from our testing. Following this table, we break down the comparisons and results and discuss how the performance advantages of Azure Database for MySQL could benefit your organization.

*Based on results from the 10-minute tests, which we did not visualize.

Achieve better performance for your money

For organizations that rely on the cloud for development or production database workloads, cloud costs may comprise a substantial line item in the budget. Keeping those costs in check must be a focus for IT.

To understand how cost might intersect with performance for the four instances we tested, we looked at a price per performance scenario. In this scenario, we examine the monthly cost to run each instance at peak performance for one hour per business day and at 25 percent of peak performance for seven hours per business day.

For the Azure instance, IOPS usage affects price, so we estimated cost with that in mind. To approximate IOPS for cost estimates, we took the average 180 million requests per hour we observed in our testing and divided by 3,600 seconds/hour to convert this into IOPS. During our tests, our Azure instance averaged approximately 50,000 IOPS. (Note some Azure pricing pages refer to both pricing per million “requests” and “IOPS” interchangeably. We confirmed with Microsoft that these terms were equivalent for pricing estimates.)

Figure 3 shows the normalized price per performance that each instance achieved at different thread counts and run times in this scenario, with lower costs being better. With Azure Database for MySQL, organizations could pay up to 54 percent less for the same performance, helping them accomplish more work while remaining in budget.

We took advantage of an accelerated logging feature for the Azure service, which is now available at no additional cost. (During our testing, the feature was available only in preview.) That feature can help boost performance and is included with Azure Database for MySQL.¹³

Note that expenses vary depending on the plan an organization chooses, which could include using the cloud for short-term performance bursts or needing longer-term options (e.g., one year, three years).

Conclusion

Leveraging a cloud service for MySQL could enhance overall operational efficiency while allowing for seamless database scaling as needed. However, when it comes to MySQL performance, Azure Database for MySQL delivered advantages over Amazon Aurora MySQL, Amazon RDS for MySQL, and Google Cloud SQL for MySQL. In our tests, Azure Database for MySQL handled more TPS with lower latency than the three competitors—up to 2.26 times the TPS and up to 55 percent lower latency—than the three competitors. The Azure service also cost up to 54 percent less for the same performance in our scenario. If the advantages of running MySQL in the cloud appeal to you, consider leveraging Azure Database for MySQL for the performance boost it could provide to your essential workloads.