At Q2BSTUDIO we are heavy users of RDS Aurora on AWS for our production workloads because it reduces the need for infrastructure maintenance and offers a very durable platform. We recently evaluated Aurora Serverless v2 to see if it fit dynamic scaling needs, and in this article we explain what Aurora is, what Aurora Serverless v2 consists of, how we tested it, the results we obtained, and why we ultimately did not adopt it for our product.
What is Aurora: Amazon Aurora is a fully managed relational database engine compatible with MySQL and PostgreSQL, designed to offer high performance, availability, and durability without the complexity of managing the underlying infrastructure.
What is Aurora Serverless v2: Aurora Serverless v2 is an on-demand, auto-scaling configuration for Amazon Aurora. Instead of fixing an instance with a certain amount of vCPU and memory, a minimum and maximum capacity threshold is defined, and Aurora automatically adjusts resources according to load using units called ACUs or Aurora Capacity Units, where 1 ACU is approximately equivalent to 2 GiB of memory with corresponding CPU and networking.
Problem we wanted to solve: Our product has high server load and heavy database activity. Maintaining performance and scaling is critical, especially in the face of unexpected events such as massive customer migrations or DDoS attacks. With the traditional RDS Aurora model, one typically provisions a larger instance than the baseline to absorb spikes, but that means underutilized resources and high fixed costs. We hoped Aurora Serverless v2 would reduce that waste while maintaining the ability to scale quickly.
Test setup: To measure behavior in real scenarios, we used a synthetics system that simulates typical authentication sessions and representative workloads. Initially we provisioned an on-demand db.r8g.8xlarge instance and subjected it to very high load. In the serverless configuration we set a minimum of 64 ACUs and a maximum of 128 ACUs. We let the serverless instance warm up as a reader and then forced a failover so it became the writer, thus observing its behavior in production.
Positive results: Functionally, Aurora Serverless v2 did what it promised. The ACUs adjusted their capacity based on load, and peaks and valleys consistent with the additional workloads we added were observed. The failover occurred just as quickly as with a pair of on-demand instances, there were no timeouts due to memory pressure, and ACU scaling was visible in the metrics.
Negative results: There were several surprises that concern us. Provisioning a large number of ACUs was slow; for example, it took around 15 minutes to provision 90 ACUs, which is too long for some sudden spikes. Additionally, once the load stabilized after the failover, baseline latency increased and we showed greater susceptibility to small response spikes, meaning the performance observed with serverless was slower and less consistent than that of our r8g instances. We suspect this may be due to previous-generation hardware or additional virtualization layers in the serverless platform that penalize latency.
Cost: The per-ACU billing model means that in sustained high-capacity scenarios, Aurora Serverless v2 can be more expensive than a well-chosen on-demand instance. In our test, a baseline of around 80 ACUs was equivalent to a higher hourly cost than a db.r8g.8xlarge. This positions Serverless v2 as a less efficient option when the regular workload already requires significant memory and CPU over extended periods.
Why it is not the right option for us right now: After analyzing performance and cost, the combination did not meet our expectations for our clients' typical workload. The slower baseline performance and slow provisioning speed during spikes were the deciding factors. Additionally, the cost per ACU in not-extremely-dynamic scenarios can exceed that of carefully sized on-demand instances.
Cases where it may make sense: Aurora Serverless v2 can be ideal for very, very dynamic workloads where the baseline usage can be reduced to a few ACUs most of the time and quickly scale up to tens or hundreds of ACUs for short periods. It also fits if the database can be offline or in a very reduced mode outside activity windows and automatic scaling is needed without human intervention.
What Q2BSTUDIO did and what we offer: At Q2BSTUDIO, as a company specialized in custom software development and custom applications, we carefully evaluate technical and economic options before adopting technologies for our clients. We are specialists in artificial intelligence, cybersecurity, AWS and Azure cloud services, business intelligence services, AI for businesses, AI agents, and Power BI. If you need a solution that combines scalable databases, custom applications, and advanced analytics with Power BI or AI agents to automate processes, at Q2BSTUDIO we design hybrid architectures and solutions optimized for performance and cost.
Practical recommendation: For teams considering Aurora Serverless v2, we recommend running representative load tests, measuring scaling times, latencies in your real workload, and modeling costs according to usage patterns. For stable workloads with predictable spikes, a provisioned latest-generation instance may be more suitable. For highly unpredictable or intermittent workloads, Serverless v2 can save costs and operations if you accept the potential latency penalty and performance variability.
Conclusion: Aurora Serverless v2 is a powerful tool and an improvement over previous versions, but it is not always the best option for environments that require low latencies and sustained consistency. At Q2BSTUDIO we continue to evaluate options and can help you choose between provisioned instances, serverless solutions, or hybrid architectures that balance performance, cost, and security. If you want to optimize your custom applications, integrate artificial intelligence, strengthen cybersecurity, or deploy AWS and Azure cloud services with business intelligence and Power BI capabilities, contact us and we will design the right solution according to your needs.
Have you tested Aurora Serverless v2 in your environment with similar workloads? Feel free to share your experience so we can all learn together
