Today at the Six Five Summit 2021, David Brown, VP of Amazon EC2, announced the “Graviton Challenge” to help companies and developers adopt the AWS-designed Graviton2-based instances more quickly. The Six Five Summit is five-day thought leadership and strategy event presented by my company, Moor Insights & Strategy, and Futurum Research.
Brown explained at the Summit that the Graviton Challenge is a new initiative with a four-day plan (customizable over several weeks) that provides developers with a step-by-step process to move their workloads from x86 instances to Graviton2-based instances. It is not that AWS doesn’t love Intel and AMD- it does; it’s just that Amazon is giving its customers more choice, and that’s always been a good thing, right?
Amazon will offer assistance in the migration if needed with resources, tips and tricks, and a Slack channel for support. At the end of the process, developers and IT can demonstrate their unique migration for a chance to win prizes including swag, a trip to AWS re:Invent 2021, and the opportunity to present on stage at the marquee conference. Getting on stage at re:Invent I am sure, would turn any developer or IT professional into a total rock star.
I think the Graviton Challenge is a clever way to show developers what is possible in just four days, with prizes as an additional incentive. So let us dig into the details of the four-day challenge (not necessarily four consecutive days).
- Day 1 – Get familiar with Graviton2 and then identify your application’s dependencies and requirements. Resources include the video “Deep dive on AWS Graviton2 processor-powered EC2 instances” and “Getting started with AWS Graviton” on GitHub.
- Day2 – Create a plan and start porting. You can use either EC2 instances with custom or AWS-provided images or containers. Both Amazon Elastic Container Service (Amazon ECS) and Amazon Elastic Kubernetes Service (EKS) support Graviton2-based instances. The Getting Started Guide on GitHub will provide information to bootstrap the application. For Java, Python, Node.js, .NET, the application might run “as-is” or need minimal changes. Other languages like C, C++, or Go will require recompilation.
- Day 3 – Test and verify functionality. Compare the performance to an x86-64 environment. AWS is standing by to help analyze and resolve any potential performance gaps.
- Day4 – Update your infrastructure to add Graviton2-based instances and deploy. Canary or blue-green deployments will redirect a portion of the traffic to the new environment.
Finally, developers can let the world know they completed the challenge by publishing a post on social media using the #ITookTheGravitonChallenge hashtag.
The AWS custom silicon journey
I always got a bit of a chuckle when I would hear “software is eating the world” and “silicon is a commodity”. As we have seen from AWS, the real magic happens when you strategically combine silicon, software and the cloud. Graviton2 is not AWS’s first custom silicon rodeo. I have been tracking the AWS custom silicon journey that started about ten years ago with the acquisition of Annapurna Labs.
AWS was first looking to accelerate storage and I/O through custom SmartNICs. This evolved into a complete virtualization layer called the AWS Nitro system. Nitro added more acceleration and offloads functions the main CPU used to run and therefore enables 100% of the main processor to be used for applications. Others in the industry refer to this as a “DPU” (Marvell and NVIDIA) and “IPU” (Intel), but it’s cool that Nitro was there first.
The Graviton CPU was announced in 2018, signaling the first general-purpose AWS CPU and, for me, that Arm had “arrived” in the cloud. Arm was big in datacenter storage, networking, and Nitro, but not general-purpose compute where developers could run their apps. In 2019, the Graviton2 was announced, and AWS says it delivers up to 40% price-performance improvements over the x86-based alternatives and prompted a new wave of customer migrations.
“Customers love Graviton2”
Brown says customer feedback has been positive based on the price-performance improvement. AWS has publicly stated up to 40% price-performance improvement, but Brown says some customers see better numbers for some workloads.
Customers have also reported that the migration is a lot simpler than expected. AWS has done an excellent job working with the ecosystem to ensure that the various libraries and ISV software packages are available. Brown said one customer migrated an entire enterprise application in just four days which formed the basis for the Graviton Challenge. This process didn’t happen overnight. I was covering Arm in the datacenter a decade ago and a lot of open-source, industry work has helped.
Unlike the first Graviton processor, Graviton2 price-performance improvements should be a reality across a broad set of general-purpose workloads such as web serving with NGINX or Apache. Graviton2 has performed well with open-source databases and is available with RDS, the relational database service.
Customers are also experimenting with compute-intensive workloads such as HPC (high-performance computing) on Graviton2 with high-performance networking up to 100 gigabits per second.
Getting started with Graviton2
The simplest way to use Graviton2 is a fully managed AWS service with Graviton2 instances. An example would be to move an open-source database to RDS using Graviton2. Another option would be to find a simple service and assign a developer for a week or two and evaluate the progress—the Graviton Challenge is the four-day blueprint to migrate to Graviton2 successfully. AWS also has a free trial available outside of the regular free trial, where you get a burstable T4g instance on Graviton2 to qualify your applications. AWS plans to keep a free trial for some time so that customers can continue to play with Graviton2 without having to incur any cost.
Let me end by saying that the one thing I appreciate about Graviton2 is that it is AWS’s processor with the features that are right for the AWS environment, based on customer feedback. There are not too many folks that can point to the same level of customization and performance. I did appreciate that AWS did not overcommit with the first-generation Graviton by suggesting a narrower set of workloads. I was a little surprised and happy to see AWS opened up with Graviton2 because you have to think less about what types of applications can migrate into the environment.
For AWS, the custom silicon journey will continue into machine inference with Inferentia, and training with Trainium. It even has custom silicon for database query acceleration with AQUA. I like the way AWS isn’t alienating Intel, AMD, or NVIDIA as I see the company many times announcing with those companies its newest processors and AWS’s corresponding instances. There’s a big pie out there for everybody—super, super exciting. Finally, I appreciate AWS breaking the news about the Graviton Challenge at our event and cannot wait to see some of the customer stories coming from that challenge and see the best on stage at AWS re:Invent.