Throughout the history of cloud computing, the term “cloud transformation” has described the process by which enterprises migrate workloads from traditional infrastructure to cloud platforms. Discussion of migration philosophies such as “lift and shift” and “refactor and replace” are time-worn. So too is the cliche of cloud providing computing and storage as a utility: immediate, on-demand infrastructure that, by merit of being remotely managed, makes the characteristics of the underlying infrastructure neither of concern nor interest.
That time has passed. Cloud platforms offer an increasingly diverse catalogue of compute options to accommodate increasingly intensive workloads. Furthermore, the way that cloud infrastructure is architected is vital to how software is built and deployed, with those changes influencing design of infrastructure in the cloud, in data centres and at the network edge.
Infrastructure as code, application containers, and cluster management — think Kubernetes — allows applications to be disaggregated from the servers that run them. These, alongside other cloud-native practices and technologies, are key to improving application performance and environmental responsiveness.
Enter Cloud-Native Practices and Cloud-Differentiated Silicon
Although infrastructure remains important, its value is made more accessible through cloud-native practices. The variety of infrastructure available in cloud platforms is more diverse than ever. In 2018, Amazon Web Services (AWS) was the first hyperscale cloud platform to introduce Arm-powered instances using its Graviton CPUs; as of July 2022, every major cloud platform now offers an alternative to x86-64 instances. Likewise, GPUs for graphics processing and video encoding and tensor processing units (TPUs) for artificial intelligence and machine learning acceleration are important resources in cloud platforms as workloads that benefit from this infrastructure become progressively mainstream.
These technologies are often exclusive to or debut on cloud platforms, and collectively represent a growing class of cloud-differentiated silicon. This includes CPUs customized for cloud platforms, GPUs, TPUs and other field-programmable gate arrays or application-specific integrated circuits designed to optimize processing of specific workloads. Among DevOps practitioners, cloud-differentiated silicon and the varied virtual machines built on it can influence the relative performance of different workloads.
AWS is quick to celebrate the cost and performance benefits of Graviton for specific workloads — as is Ampere, which designed the Arm-based Altra processors available on Google Cloud Platform, Microsoft Azure and Oracle Cloud Infrastructure. Porting applications from x86-64 to Arm instances, however, requires developer effort. Many open-source applications now have versions optimized for Arm, and many commercial providers are plotting Arm support in their product road maps. The availability of Arm CPUs on cloud platforms is instrumental to fostering the software ecosystem, resolving the stagnation that has limited the impact of Arm processors in data centres.
Managing this catalogue of options practically requires cloud-native practices. For example, Kubernetes can aid in scheduling an appropriate amount of pods for a given host; tools like Terraform and Chef can assist in identifying dependencies on x86-64 instances for enterprises approaching a migration to Arm; and Terraform is useful for ensuring that instances are configured to provide the needed resources for an application, including different ratios of computing and memory, attached non-volatile memory storage or computing accelerators such as GPUs or TPUs.
Using these cloud-native practices in your organization’s application life cycle management will make it easier to adopt cloud-differentiated silicon and extract the full value it promises.
Considering Environmental Responsiveness in Cloud Computing
Cost is a major motivator for cloud-differentiated silicon. For the user, cheaper computing with comparable core counts, memory allocation and performance can be sufficient motivation to migrate workloads. For the cloud platform operator, designing CPUs for their own specifications and paying a foundry such as TSMC to manufacture them is cheaper than buying from Intel or AMD, given the quantities involved at a hyperscale level.
Power consumption is another consideration. Building a CPU with the assumption that squeezing every drop of performance from each core on a system is imperative, because customers are paying for software licences on a per-core basis, is starkly different to building for parallel processing. If you’re deploying open-source or internally developed cloud-native software, a higher number of power-efficient cores may be a better fit for your workload than a smaller number of powerful cores.
That power efficiency in aggregate makes a difference — the thermal design point of Intel’s Sapphire Rapids Xeon CPUs is estimated to be almost double that of AWS’ Graviton3. For AWS, transitioning managed customer workloads such as remote desktop services to Graviton reduces power consumption, which benefits its environmental sustainability targets. For enterprises opting to track performance against environmental, social and governance goals — which may soon be a requirement, pending proposed action by the US Securities and Exchange Commission — the use of cloud platforms contributes to the Scope 3 emissions of your organization as categorized by the Greenhouse Gas Protocol.
Although cloud-differentiated silicon can help cut emissions by using power-efficient processors and workload-specific accelerators, a lot more work is needed to improve the tracking and reporting of cloud emissions. Highlighting environmental data such as the power usage effectiveness of a cloud region, the mix of carbon and carbon-free energy and net operational greenhouse gas emissions, alongside pricing explanations in monthly statements is a good first step. Exposing this information through an API to enable integrations with third-party services would be transformative.
In the same way that cloud management platforms were designed to help enterprises better control costs in hybrid and multicloud systems, providing environmental data through an API could prompt the creation of emission management platforms. This would enable enterprises to make informed choices about the environmental impact of cloud deployment strategies, consolidate reporting across environments and assist in reporting compliance.
Where Do We Go from Here?
Cloud platform operators have communicated messages of faster, cheaper and more secure cloud-differentiated silicon, but the impact of these options for developers would benefit from further articulation. Deeper exploration of this impact will be a recurring point of collaboration between the Cloud and Infrastructure and Developer Trends practices at CCS Insight — this theme is likely to be a highlight in our reports on Microsoft Ignite and Google Cloud Next this October and at AWS re:Invent in early December.
Likewise, environmental, social and governance implications will be explored in greater detail next year, as cloud platform operators intensify conservation efforts and amplify environmental messaging in April 2023, coinciding with Earth Day.