Why hardware matters for the cloud?

September 19, 2019

A blog post by Doug Bourgeois, managing director, Deloitte Consulting LLP, Sean Vandruff, specialist leader, Deloitte Consulting LLP, Taylor Jones, business technology analyst, Deloitte Consulting LLP and Zachary Zweig, business analyst, Deloitte Consulting LLP

Cloud is a crucial part of the conversation for any organization committed to maintaining a modern, robust IT infrastructure. That being said, the transition to the cloud is more than turning on a simple switch. Organizations must make technical decisions driven by strategic actions, ensure that applications run securely with optimized performance, and work to achieve cost benefits.

However, hardware considerations are often overlooked when deploying a public, private, hybrid, or multicloud. Wherever an organization is in its cloud migration journey, it is crucial that organizations identify the underlying hardware infrastructure and management frameworks that will provide mission-critical capabilities at minimum cost.¹This is particularly the case for hybrid and multiclouds. These architectures are gaining in popularity—especially in the government and public sector, in part due to the unique security and regulatory compliance requirements of such organizations. If properly created and managed, hybrid cloud and multiclouds operate well at scale and during transition, enabling large organizations to roll out cloud migration in a way that minimizes risk and maximizes overall adoption.

Hardware matters for any cloud model
Technology companies and hardware manufacturers, like Intel^®, have successfully optimized software, architectures, and hardware for the complex level of computing required to run next generation workloads including artificial intelligence (AI) and edge computing. This optimization allows organizations to pursue cloud models tailored to their specific mission needs. In some use cases, raw computing power is paramount to success and the total cost of the solution will reflect that decision. However, due to budget constraints, most applications need to be optimized based on value, i.e., the best performance per watt per dollar for your specific mission capability needs.

This value-based cost optimization is complex, nuanced, and mission-specific. For the on-prem components of a hybrid cloud model, cost-sensitive organizations can get the most “bang for their buck” by purchasing carefully optimized hardware stacks at a reasonable price as opposed to either overly powerful–and exceedingly expensive–flagship hardware or cheaper and less reliable commodity hardware. Once hardware decisions have been made, an organization can ensure developers are deploying to the most cost-effective machines without sacrificing performance.

Hardware considerations must also be taken into account for services deployed to public clouds as well. For example, an organization may design computer vision applications that rely on the significant processing, performance, and power usage improvements of Field Programmable Gate Arrays (FPGAs) over standard CPUs. As such, if the same application is to be migrated to a public cloud environment, it must be provisioned to cloud resources that support FPGAs, a functionality not integrated into all Cloud Service Provider (CSP) architectures.²

Similar considerations should be provided for other applications designed for intelligent processing of video, voice and natural language, or similar advanced workloads leveraging visual processing, artificial intelligence, machine learning, and other cognitive tasks.³ These technological capabilities are rapidly increasing in usage as the journey to the cloud continues to progress. As a result, organizations may not yet have experience with the optimization of these workloads, including addressing the intense requirements that they can place on the underlying hardware. This means that the need to carefully consider the capabilities and services provided by the hardware with such workloads is still necessary to achieve the most cost-effective solution without sacrificing performance–even in the cloud.

According to a 2018 survey of nearly 1,000 tech executives, more than 80 percent of organizations are deploying a multicloud strategy with hybrid models being the most common. These hybrid cloud architectures of the future will be driven by the requirements of next generation workloads and the need to address risks, threats, compliance, and governance. As a result, underlying infrastructure and management frameworks cannot be treated as a commodity—they require informed, conscious decisions about the types of hardware systems being run in the cloud and on the edge.

Endnotes

¹HCRA (The hybrid cloud architecture of the future must be driven by the missions, business outcomes, and the characteristics of next generation workloads such as AI, ML, and Deep Learning neural networks, combined with increasing back-office requirements for addressing risks, threats, compliance, and governance. As a result, underlying infrastructure and management frameworks cannot be treated as a commodity, but rather require informed, conscious decisions about the types of hardware systems are being run on and the types of public cloud instances organizations are leveraging.)

Bourgeois, Douglas; Vandruff, Sean. "Designing Hybrid Cloud Architecture for the Future." Deloitte Consulting LLP, LLC., (September 2019)

²HCRA (However, for successful cloud adoption, private cloud workloads leveraging these architectural features must be migrated to cloud resources which support the same features. For example, machine vision application developers may design applications relying on the significant processing, performance, and power usage improvements of Field Programmable Gate Arrays (FPGAs) over standard CPUs. As such, the same application migrated to a cloud environment should also be provisioned to cloud resources which also support FPGAs.)

Bourgeois, Douglas; Vandruff, Sean. "Designing Hybrid Cloud Architecture for the Future." Deloitte Consulting LLP., (September 2019)

³HCRA (Similar considerations should be provided for other applications designed for intelligent processing of video, voice and natural language, or similar advanced workloads leveraging VPUs, AI/ML, and other Cognitive tasks.)

Bourgeois, Douglas; Vandruff, Sean. "Designing Hybrid Cloud Architecture for the Future." Deloitte Consulting LLP., (September 2019)