What to Consider When Choosing AWS Instances

Moving to the cloud can help reduce the size and frequency of large IT investments. But the cost of running the business day-to-day often goes up. In fact, cloud spending is exceptionally high for many businesses. By one estimate, 45% of all enterprise IT budgets will be spent on the cloud by 2026.(1) Cloud users themselves estimate that 30% of their spend is wasted.(2)

Everyone is under pressure to cut their cloud spend—and there are many ways to manage these costs. Examples include using reserved instances, spot instances, and savings plans; modernizing code; going serverless; and eliminating licensing costs by reducing cores and adopting open-source applications. Other approaches, such as choosing instances based on published hourly rates, can impact operations.

Choosing instances with the newest generation of processors, however, can increase both overall performance and price performance to help reduce cost. Many Amazon Web Services (AWS) users are running workloads on AWS instance generation 4 or 5, but AWS is on instance generation 7 now and is expected to announce instance generation 8 later in 2023. Maybe those numbers don’t sound all that far apart, but it’s worth remembering that instance generation 4 went GA (general availability) in 2015, and instance generation 5 went GA in 2016.(3) That’s not only several years ago, but also several generations of processors ago. What difference does that make to users?

Why Choose a Newer Instance Generation?

Performance increases on more advanced technology, and hardware efficiencies translate into better performance per dollar, which can help reduce your overall cloud spend. That’s not all—just the act of moving from one cloud instance to another can result in efficiency improvements. The things you do to move, like changing operating systems, updating software versions, confirming SLAs, assessing provisioning, and cleaning code, all add up to better outcomes.

Why Choose a Newer Instance Generation with Intel?

Moving to newer-generation instances with the same microprocessor architecture can pose fewer challenges than moving to instances based on an entirely different microarchitecture. An example is moving from C5 instances with Intel to C6i or C7i instances with Intel. Beyond that, there are other considerations around the underlying hardware and the cloud to keep in mind.

Hardware Considerations

• Memory Performance:
  • Newer instance generations are more likely to use different types of memory or offer better memory performance that’s in line with technology changes made after older instance generations went GA.
  • Better memory boosts applications that need responsive, low-latency memory performance, such as large in-memory databases or high-performance computing (HPC).
• Start-up and Initialization Times:
  • Newer instance generations often start-up or initialize faster than older instance generations due to improved hardware or rack optimizations.
  • Faster initialization time can benefit workloads that are sensitive to startup times. A good example is web servers, which need to spin up quickly on demand.
• Custom Kernels:
  • Custom kernels require validation for compatibility with the newer-generation instance.
  • Hardware or virtualization changes sometimes require kernel adjustments.
• Accelerators:
  • Workload performance improves with publicly available configurations and libraries that turn on built-in accelerators in AWS 6i and 7i instances—with no additional fees.
  • Common applications benefit from on-processor capabilities for tasks like encryption, compression, and networking.
    • With acceleration enabled, C6i delivers 1.50x more WordPress transactions per second than C6g, R6i delivers 8.45x more connections per second in NGINX than R6g, and M6i has 8x higher inference throughput than M6g ResNet-50 image processing.(4)
    • AI functions speed up even more in 7i instances. Natural Language Processing (NLP) using the RoBERTa model ran up to 10.65x faster on M7i than on M7g.(5)
  • Architects may be tempted to dismiss accelerators as an academic exercise for developers, but choosing cloud instances that give coders these built-in options saves operational expenses in two key ways.
    • First, with better performance, you can provision smaller instance sizes at lower rates or spend less time running your workloads on larger instances.
    • Second, the dollars you spend simply go farther. Price performance for the applications in the bullet points above is 1.20x to 8.62x better on newer-generation Intel than on comparable Graviton instances, and that’s just rental cost.(6)

AWS Considerations

• Underlying Platform Changes:
  • Changes that AWS makes in its server racks may not be top of mind for a cloud user, but they can have impacts on performance, security, and compatibility.
  • AWS has changed underlying virtualization technology from one instance generation to the next (for example, from Xen to Nitro).
• Elastic Network Interfaces (ENIs) and Network Performance:
  • Each instance generation might support a different number of ENIs or offer different network performance.
  • By optimizing your configuration, ENIs can help increase workload performance on the newer instance generation.
• Amazon Elastic Block Storage (EBS) Optimization:
  • Newer-generation instances often come with improved Amazon EBS–optimized capabilities.
  • These capabilities can potentially offer higher throughput or input/output operations per second (IOPS).
• Instance Metadata:
  • AWS introduced Instance Metadata Service Version 2 (IMDSv2) as a defense-in-depth measure against open firewalls, reverse proxies, and Server-Side Request Forgery (SSRF) vulnerabilities.
  • It requires a session token to access instance metadata (obtained via HTTP PUT request).
  • Ensure that your applications and tools that rely on instance metadata are compatible with any changes to the instance they run on.
• Burst Performance and Credits:
  • Burstable performance is closely associated with instance generation.
  • AWS may adjust the CPU credit model or rates between generations.

To learn more about the architectural nomenclature of AWS instances and how to pick the right one to meet your business needs, see the report from Prowess Consulting: "Cloud Computing: Why You Should Be Looking Under the Hood."

Why Choose Intel Instances Instead of Graviton?

I have been working with enterprise deployments on-premises and in the cloud for over 20 years. What I am looking for, and you probably are too, is the best infrastructure for operations, for the best value, with a simple, reliable set up that gives my team control over making changes when we need to. Basically, IT has to work well today and be nimble tomorrow for a reasonable cost. How do Graviton and Intel compare in achieving those goals?

Graviton is exclusive to the AWS cloud and offered only as a managed service.(7) Each Graviton processor generation is built on unique combinations of licensed IP and proprietary AWS design. The first Graviton processor is different from Graviton2, which is different from Graviton3. Not all software that runs on Intel will run on Graviton processors without modification, and not all software that runs on Graviton2 will run on Graviton3 without additional time and effort, and not all software that has been optimized for one Graviton generation will run efficiently on another proprietary processor with similar licensed IP—those processors are all unique by design. If you are one of the 72% of companies that have a hybrid cloud strategy,(8) locking your business into a one-off technology like Graviton doesn’t make sense.

On the other hand, as of mid-2023, Intel had shipped 85 million generationally compatible Intel Xeon processors.(9) No effort is required to re-acquire baseline functionality to move among Intel instances. Intel delivers consistent performance for your workloads in 50,000 instance types from leading CSPs,(10) as well as private clouds, edge sites, and MSPs. Ninety percent (yes, 90%) of developers are using software developed or optimized by Intel.(11) You can run your workloads on Intel with dozens of public and private cloud providers or build an on-premises data center with Intel. Choosing Intel keeps your options open to change deployment models and establish relationships with any number of providers and systems integrators whenever it makes sense for your business.(12)

At a more granular level, the amount of work that can go into refactoring code to run your production workloads on a Graviton instance can range from recompiling the code and its dependencies to rewriting entire applications from the ground up. Container images are also architecture-specific and will need to be rebuilt. Beyond demanding resources, switching to Graviton can have other ramifications:

• Libraries and dependencies: The libraries and dependencies used by your applications might not be available and optimized for the Graviton processor in a given instance.
• Middleware and runtime: Middleware or runtimes like Java or .NET Core require optimization, validation, and testing for the Graviton processor in a given instance.
• Driver and kernel module compatibility: Specific drivers or kernel modules might not be compatible with or have an equivalent alternative to the Graviton processor in a given instance.
• Amazon Machine Images (AMIs) and bootstrap scripts: Any bootstrap scripts or configurations in your AMIs that make architecture-specific assumptions will need to be reviewed and updated for the Graviton processor in a given instance.
• Database compatibility: Your database software might not be compatible with the Graviton processor in a given instance, its performance may not meet your service-level agreements (SLAs), and/or there may be changes to your licensing costs.
• Third-party software: Applications that rely on third-party vendors might not offer versions of their software compatible with all Graviton generations, or they may provide fewer options and less support and services for Graviton than they do for other processors.

Not all dependencies stem from software; some of your organizational practices might need to be evaluated or changed to run on instances with a Graviton processor:

• Build and continuous implementation/continuous deployment (CI/CD) adjustments: Your build and CI/CD pipelines might be redirected or split to support Graviton, and perhaps to support different generations of Graviton. This is especially true if you compile code as part of your build pipeline.
• Debugging and troubleshooting: Tools, utilities, or practices you use to debug and troubleshoot might not directly apply to the Graviton processor, or work for all generations of Graviton. You might have to re-tool and acquire different talent to maintain and troubleshoot multiple development environments.
• Training: Your development, operations, and support teams might require additional training. That takes time and is narrowly useful as Graviton is only available in the AWS cloud, not from other CSPs or through an SI for on-premises data centers.

Looking at all of those factors together, how do Graviton and Intel compare? Intel works today, is nimble tomorrow, and has a reasonable cost. Graviton doesn’t even come close.

Choose to Save Money

Many organizations—and individuals within those organizations—are under pressure to reduce their cloud spend. AWS recommends Graviton to cut cloud costs, but any savings on the monthly bill are likely the result of choosing a newer instance generation, the coding work you did to move your workloads, and discounted services from the CSP—not something inherent in the Graviton processor. Instead, plan today to keep your level of effort low to develop and maintain your workloads, keep your costs transparent, and keep your options open by choosing newer instance generations with Intel.

If you are an AWS user in instance generation 4 or 5, consider choosing newer-generation instances with Intel to take advantage of newer technology and manage your cloud spend effectively without adding complexity and uncertainty to critical operations in the cloud. You may be looking at instance generation 6 with Graviton2 for the low hourly rate, and it’s wise to compare it with Intel in the same instance generation. Intel can help you reach your goals in the AWS cloud. Visit intel.com/cloudfacts and contact us.

--------------------------------------------------------

(1) Zippia. “25 Amazing Cloud Adoption Statistics [2023]: Cloud Migration, Computing, And More.” June 2022. zippia.com/advice/cloud-adoption-statistics/.

(2) “In 2020, respondents report that 30 percent of cloud spend is wasted. Due to this finding, many organizations consider cloud cost management a top priority.” Source: Statista. “Percentage of cloud computing spend wasted in organizations worldwide in 2020.” March 2023. statista.com/statistics/1252987/cloud-spend-wasted-vs-efficient-spend/.

(3) Jeff Barr, Chief Evangelist for AWS. “EC2 Instance History.” May 2015. https://aws.amazon.com/blogs/aws/ec2-instance-history/.

(4) 6i instances have Intel Crypto Acceleration, Intel QuickAssist Technology Engine for OpenSSL, and Intel Deep Learning Boost. Source: Intel. “Accelerate Cloud Work in AWS Instances,” August 2023. https://www.intel.com/content/www/us/en/content-details/769257/accelerate-cloud-work-in-aws-intel-instances.html

(5) 7i instances have the same acceleration capabilities as 6i and more, including Advanced Matrix Extensions (Intel AMX) with 8-bit integer, bfloat16, and float16 support used in RoBERTa testing. Source: Principled Technologies, sponsored by Intel. “Accelerate natural language processing with AWS EC2 M7i instances featuring 4th Gen Intel Xeon Scalable Processors.” October 2023. https://facts.pt/RfrK3Rr and https://facts.pt/2aqfhRt.

(6) Intel. “Accelerate Cloud Work in AWS Instances.” August 2023. https://www.intel.com/content/www/us/en/content-details/769257/accelerate-cloud-work-in-aws-intel-instances.html. Additional source: Principled Technologies, sponsored by Intel “Accelerate natural language processing with AWS EC2 M7i instances featuring 4th Gen Intel Xeon Scalable Processors.” October 2023. https://facts.pt/RfrK3Rr and https://facts.pt/2aqfhRt.

(7) See https://aws.amazon.com/outposts/. An on-premise option is a managed service; there is no option to independently own/purchase servers with Graviton processors.

(8) Flexera. “2023 State of the Cloud Report.” 2023. https://info.flexera.com/CM-REPORT-State-of-the-Cloud.

(9) See https://www.intel.com/content/www/us/en/content-details/768566/4th-gen-intel-xeon-scalable-processors-30-3-30-public.html?wapkw=spr&DocID=768566

(10) Liftr. “Liftr Cloud Components Tracker.” April 2021. https://liftrinsights.com/cloud-components-tracker, cited on https://intel.com/cloudperformance.

(11) See https://www.intel.com/content/www/us/en/content-details/768566/4th-gen-intel-xeon-scalable-processors-30-3-30-public.html?wapkw=spr&DocID=768566, from Global Development Survey conducted by Evans Data Corp., 2021.

(12) “Liftr Insights data show that not only did AWS increase their average prices in 2020, 2021, and 2022, but the increases have been higher each year since 2019.” See https://www.prnewswire.com/news-releases/aws-and-azure-cloud-pricing-moving-in-different-directions-as-shown-by-liftr-insights-data-301736920.html

Notices and Disclaimers

Performance varies by use, configuration, and other factors. Learn more on the Performance Index site.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available ​updates. See backup for configuration details. No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software, or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.