There’s a new player in town: generative AI (GenAI). This technology thrives on swift data collection, training and inferencing. Consequently, optimizing your AI applications — and, by extension, other high-value workloads — hinges on the velocity, scalability and efficacy of your infrastructure and the maturity of your DevOps processes.
Achieving maturity in a DevOps organization requires overcoming various barriers to attain the short- and long-term goals set for the infrastructure. Short-term goals include upskilling the IT team’s resources and integrating containerization and automation tools that are used throughout the operating lifecycles (from Day 0 to Day 2). By scaling up containerized environments and automating processes, you can significantly enhance your company’s long-term economic viability and sustainability. Long-term goals involve deploying these solutions across multicloud and multisite landscapes and balancing workloads.
A survey conducted shortly before the AI explosion revealed that the level of automation in infrastructure operations’ workflows was generally below 50% among companies navigating the modern data landscape. Coupled with a twofold increase in application counts, organizations may find themselves struggling to keep pace with the relentless tide of change.
GenAI Infrastructure: Easy as Cake?
From compute capabilities to storage density and speed, spanning unstructured, block and file formats, fundamental elements of automation are ripe for swift integration of automation across the organization to establish a robust foundation for GenAI implementation. A smart way to ease the journey toward AI is by layering prebuilt integration tools into a portfolio of products. It’s like creating a multi-tiered cake.
There are important considerations when selecting the various hardware infrastructure components for a generative AI system, including high-performance computing, high-speed networking and scalable, high-capacity and low-latency storage. The infrastructure requirements for AI and machine learning (ML) workloads are dynamic, and depend on factors including the nature of the task, the size of the data set, the complexity of the model and the desired performance levels. There is no one-size-fits-all solution for GenAI infrastructure, as different tasks and projects may demand different configurations.
Infrastructure as Code for AI
Infrastructure as Code (IaC) principles, which facilitate the automation and orchestration of underlying infrastructure components, are central to the success of generative AI initiatives. By leveraging IaC tools like Red Hat Ansible and HashiCorp Terraform, organizations can streamline the deployment and management of hardware resources, ensuring seamless integration with GenAI workloads.
These popular IaC tools improve existing high-value workloads and simplify the process of onboarding new AI application workloads. For example, they enable you to craft playbooks and plans for automating server configurations, provisioning, deployments and updates while data collection is underway.
Ansible is versatile, supporting imperative and declarative programming. Due to their mutable nature, playbooks can be modified in real time without disruption to ongoing processes or end users. Ansible modules, playbooks and roles can be engineered for the compute configurations necessary to meet the unique demands of large language models (LLMs) and workloads. Terraform offers a declarative approach to environment setup and teardown, which is ideal for scenarios where repeatability and consistency are paramount. These two dominant IaC tools have different functional purposes in automation across IT ops; there is no one or the other.
Storage for AI
AI data processing and training require access to scalable and simple file systems, particularly as training data increases. AI also demands unstructured data storage to access the bounty of rich context and nuance during the building phase. Variable user access is another AI requirement, and Ansible automation playbooks can enable this quick change and adaptation.
End users realize AI’s true benefits when they actively engage in real-time with inferenced data drawn from trained AI models via GenAI tools like Copilot, ChatGPT or Dall-E. These applications may reside on-premises or in the cloud, tapping into data from block or file storage arrays that offer low latency and cost-effective processing. Making such arrays available is pivotal to delivering GenAI outputs that meet end users’ needs.
With automation firmly in place, the infrastructure furnishes the elasticity necessary to propel movement, learning and the evolution of generative AI. For example, Terraform is excellent for provisioning cloud infrastructures and tearing them down when they’re no longer needed. This can be useful when migrating workloads from one place to another or when testing and forecasting how much storage is needed for a GenAI workload. And as for post-production, you can use Ansible for provisioning or expanding target storage without disrupting activities in motion.
Bake an AI-Ready Infrastructure with IaC
Generative AI tools need greater scale, repeatability and reliability than anything previously created by combining software and data centers. This is also precisely what building Infrastructure-as-Code practices into a multisite operation is designated to do. It’s like a layered cake: Without the ingredients baked in properly, you can’t appreciate the frosting on top.
The post GenAI Acceleration Depends on Infrastructure as Code appeared first on The New Stack.
By facilitating the automation and orchestration of infrastructure components, IaC is central to the success of generative AI.