AI Infrastructure: Storage Options

Overview

In this course, you’ll take a comprehensive journey through the storage solutions available on Google Cloud, specifically tailored for AI and high-performance computing (HPC) workloads. You’ll learn how to choose the right storage for each stage of the ML lifecycle. You’ll explore how to optimize for I/O performance during training, manage massive datasets for data preparation, and serve model artifacts with low latency. Through practical examples and demonstrations, you’ll gain the expertise to design robust storage solutions that accelerate your AI innovation.

Syllabus

Course overview

This module offers an overview of the course and outlines the learning objectives.

Foundations of AI storage

This module details the role of storage infrastructure in the AI data pipeline. It covers performance demands, key Google Cloud solutions, and the decision criteria for selecting a service based on capacity, throughput, and latency.

Prepare and train

This module details the critical phases of data preparation and model training within the AI workflow. It covers optimizing data loading using Cloud Storage, Anywhere Cache, and the Dataflux Dataset tool, while comparing high-performance file systems like Cloud Storage FUSE and Managed Lustre. Additionally, it outlines decision criteria for efficient checkpointing strategies to ensure fault tolerance and minimize GPU idle time.

Serve and archive

This module details strategies for AI model serving and data archiving. It covers selecting storage—Managed Lustre, Cloud Storage, or Hyperdisk ML—based on scale and latency, and optimization techniques, like GKE Image Streaming and Cloud Storage FUSE, to minimize costs and load times.