AWS Graviton & Traditional x86 Instances: Performance Analysis

AWS Graviton is a 64-bit ARM-based CPU designed by AWS for their cloud instances/resources and launched first at Re-invent 2018. Graviton CPUs are not available outside of the world as it is used within the AWS Data centers only.

Traditionally, we used x86 AMD-based CPU for our workloads, which is designed and developed by Intel and has a good share in the market nowadays, but Now, with Graviton CPUs, there are significant optimization differences that help us to reduce the Cost, Performance, etc. 

Below is a short introduction about the  x86 and Graviton Instances about their nature, workload support, request handling, etc. : 

●  x86 instances: x86 CPU Instances are AMD-based architecture, which is a good fit to run our general workloads like Web servers, Microservices, Containerization, and small, scalable applications. These instances provide a variety of workload support like enterprise application Content Management Service application, and it is also ideal for Database workloads due to its high core and memory. It can handle multiple concurrent requests and queries at a time with good performance.

● Graviton Instances: Graviton instances are ARM-based custom-designed instances provided by AWS. These are 40% cheaper than traditional x86 instances. These are the alternate solutions of x86-based instances. Graviton instances are a good fit for power and cost-efficient workloads. ARM-based processors are commonly used in Mobiles, and we are aware of their performance and scalability. Graviton-based instances provide a variety of instance options that are optimized for various workloads.

As we see above, the overview of both x86 and Graviton instances and when we start our Graviton migration journey, we have multiple concerns about efficiency, scalability performance, etc. We want to ensure that our workload runs with good performance as well as minimal cost. 

So, keeping things in mind about all concerns and performance queries, below are some of the points that help us to understand Graviton Instances : 

● vCPU:

● Graviton: Graviton processors showed good performance per vCPU core when there are larger threads assigned on the CPU core accordingly. It gives 10-15% advantages over the Intel(x86) Processors, even in high concurrency scenarios. Graviton advantage could be 70000 transactions per second over the Intel traditional x86 processors in real numbers.

● x86: Traditional x86 processors showed fast performance when it comes to a single core handling the single threads or the smaller number of threads assigned accordingly. It has better performance with the low concurrency scenarios, and when it has some additional free vCPU, it could be better and has advantages up to 35%.

● Price to Performance: 

● Graviton: These instances are generally more cost-effective than the traditional x86 processors. Aws provides better prices for Graviton instances as compared to the traditional x86 processors for price-conscious customers. Graviton processors provide 15-25% better price-performance for disk-bound and CPU-bound workloads.

● x86: Traditional x86 instances are quite expensive compared to Graviton instances, and when it comes to the single-threaded workloads, x86 instances offer 6% and 14% better price to performance as compared to Graviton instances. When all CPU resources are fully utilized, and spot instances are being used, then high-clocked Intel (x86) instances provide 20% and 8% better price performance compared to the Graviton

● Memory and Storage: 

● Graviton : 

● Memory: Graviton instances come with various configurations of memory, and it does not always match the highest configuration provided by traditional x86 processors but it is optimized for several use cases, and you can refer accordingly to your workloads.

● Storage: Graviton instances support EBS volumes and instance store volumes for their storage requirements.

● x86 : 

● Memory: x86 typically offers a variety of memory options from the range of Gigabytes to Terabytes depending on instance size. Instances are designed for Memory-oriented workloads. R-series and X-series instances offered higher memory.

● Storage: x86 instances also provide various storage options like Graviton. We can use EBS as well as an Instance store to fulfill the requirements of storage.

Workload Compatibility:

● Graviton: Workload compatibility will be the reason for consideration because Graviton instances are optimized for various types of workloads, so it gives us the flexibility to choose wisely. It is compatible with several types of applications, databases, etc.

● x86 : Traditionally, x86 is also better optimized for some applications due to dominant market share from the last few years.

● Software Ecosystem: 

● Graviton: Since Graviton is quite new in the market nowadays, a lot of libraries and packages are built that support Graviton, but there are still a few packages and libraries that might not have native support for ARM chipsets.

● x86: Traditional instances have a well-established software ecosystem due to its long market shares. It provides better software packages compared to Graviton.

Here is the quick comparison table of the above parameters to get the context of both instances according to their specific workloads.

Conclusion:

Graviton instances are significant in performance and cost-effectiveness, challenging the traditional x86 instances in certain scenarios. It also provides a wide range of optimization for several workloads, which helps us to reduce the cost of our resources along with good performance and compatibility. It has a good price-performance ratio compared to traditional x86 processor-based instances according to the industry’s workloads. We can keep the above few aspects while analyzing the performance of conventional x86 and Graviton instances.

References: 

https://aws.amazon.com/ec2/graviton/

https://attriab.medium.com/why-aws-graviton-is-a-game-changer-for-cost-effective-computing-80e8dcbfa71c

https://www.percona.com/blog/comparing-graviton-performance-to-arm-and-intel-for-mysql/