AWS Compute Blog

Performance benefits of new Amazon EC2 R8a memory-optimized instances

Recently we announced the availability of Amazon Elastic Compute Cloud (Amazon EC2) R8a instances, the latest addition to the AMD memory-optimized instance family. These instances are powered by the 5th Generation AMD EPYC (codename Turin) processors with a maximum frequency of 4.5 GHz. In this post I take these instances for a spin and benchmark MySQL later on, but first I discuss the top things you should know about these instances.

Notable characteristics of R8a instances

Each vCPU on an R8a instance corresponds to a physical CPU core (something we started on 7th generation AMD instances). This means that there is no simultaneous multi-threading (SMT). Each vCPU mapped to a dedicated physical core, which means that you get more predictable and consistent performance because there’s no resource sharing or potential interference between threads, which is particularly crucial for performance-sensitive workloads where consistent latency is essential. When evaluating and adopting R8a instances, make sure that you’re re-evaluating your thresholds for CPU usage. You can likely squeeze more out of each instance’s CPU without impacting any of your workload’s SLA metrics.

R8a instances feature sizes of up to 192 vCPU with 1,536 GiB RAM. The following table shows the detailed specs:

Instance size vCPU Memory (GiB) Instance storage Network bandwidth (Gbps) EBS bandwidth (Gbps)
r8a.medium 1 8 EBS Only Up to 12.5 Up to 10
r8a.large 2 16 EBS Only Up to 12.5 Up to 10
r8a.xlarge 4 32 EBS Only Up to 12.5 Up to 10
r8a.2xlarge 8 64 EBS Only Up to 15 Up to 10
r8a.4xlarge 16 128 EBS Only Up to 15 Up to 10
r8a.8xlarge 32 256 EBS Only 15 10
r8a.12xlarge 48 384 EBS Only 22.5 15
r8a.16xlarge 64 512 EBS Only 30 20
r8a.24xlarge 96 768 EBS Only 40 30
r8a.48xlarge 192 1536 EBS Only 75 60
r8a.metal-24xl 96 768 EBS Only 40 30
r8a.metal-48xl 192 1536 EBS Only 75 60

Testing MySQL performance using HammerDB

R8a instances are a great choice for MySQL databases, so I thought that would be a great place to showcase some of these instances capabilities. To test MySQL, I used a series of scripts written by my colleagues to track MySQL performance across software versions and different EC2 instances. These scripts are stored in the repro-collection repository, which is an open source, extensible framework for performance testing that addresses real-world workloads rather than micro-benchmarks. It is built to provide a performance measurement reference usable across multiple organizations, and it’s currently centered on MySQL and actively used in discussions with Linux Kernel developers and maintainers. Furthermore, it helps track any performance impacts created by code changes to MySQL. The scripts contained in this repository set up a MySQL database to be tested, and a load generator running the HammerDB benchmark.

For this benchmark I used an r6a.24xlarge instance for the load generator, and an r6a.xlarge, r7a.xlarge, and r8a.xlarge instances for the MySQL database server all deployed in the same AWS Availability Zone (AZ). I chose a single AZ setup to minimize any latency variability from crossing multiple AZs. This is not meant to be a production-like setup, and I highly recommend using multiple AZs for production workloads. Each MySQL instance was tested separately using the same HammerDB load generator. Each test was run three times, and the results were averaged across the three runs. A diagram of the architecture is shown in the following figure:

Performance testing architecture showing r6a/r7a/r8a instance types with HammerDB load generator executing 9 test runs

HammerDB overall results

R8a instances show great results in the HammerDB benchmark for MySQL databases. For HammerDB’s overall score category, R8a instances outscored R7a instances by 55% and outscored R6a instances by 74%.

Performance comparison chart showing r6a, r7a, and r8a instance scores

HammerDB transactions per minute test

R8a instances also showed a notable improvement in this category. When compared to previous generation R7a instances, R8a out performed R7a by 32%. When compared to R6a instances, R8a outperformed by 63%.

Performance comparison showing r6a (91,105), r7a (112,686), and r8a (148,478) transactions per minute

HammerDB P99 latency results

R8a instances showed improvement in P99 latency results, showing the efficiency gains driven by the new 5th Generation AMD EPYC CPUs and higher memory bandwidth. R8a shows an 14% latency reduction when compared to R7a, and a 25% latency reduction when compared to R6a.

P99 latency comparison showing decrease from 39.93ms (r6a) to 30.02ms (r8a) across instance generations

Conclusion

Built on the AWS Nitro System using sixth generation Nitro Cards, R8a instances are ideal for high performance, memory-intensive workloads, such as SQL and NoSQL databases, as demonstrated by the bench-marking shown in this post, as well as distributed web scale in-memory caches, in-memory databases, real-time big data analytics, and Electronic Design Automation (EDA) applications. R8a instances offer 12 sizes, including 2 bare metal sizes. Amazon EC2 R8a instances are SAP-certified, and providing 38% more SAPS when compared to R7a instances. If you’re still running 6th generation R6a instances, then I highly encourage you to migrate to the 8th generation instances to use their clear price performance benefits. Staying on modern infrastructure is a great way to drive down costs and provide more features for your customers, and there are clear gains to be had based on the testing shown in this post.

Start optimizing your high performance memory intensive workloads today by migrating to R8a instances. Visit the Amazon EC2 R8a instances page to learn more and get started on your upgrades to use the increased price performance of R8a instances today!