This document provides an overview of AI zones for Cloud Storage. AI zones are specialized Google Cloud zones that are designed to offer computing capacity for artificial intelligence (AI) and machine learning (ML) workloads. They provide significant ML accelerator (GPU and TPU) capacity.
AI zones are optimized for AI and ML workloads like the following:
- Large-scale training
- Small-scale training, fine-tuning, bulk inference, and retraining
- Real-time ML inference
For background information about AI zones, see AI zones in the Compute Engine documentation.
Within a region, AI zones might be geographically located away from standard (non-AI) zones.
AI zones are compatible with other Cloud Storage and Google Cloud features.
Storage architecture recommendations
We recommend that you use a tiered storage architecture to balance cost, durability, and performance:
-
Cold storage layer: use regional Cloud Storage buckets in standard zones for persistent, highly durable storage (the "source of truth") of your training datasets and model checkpoints.
-
Performance layer: use specialized zonal storage services to act as a high-speed cache or temporary scratch space. This approach eliminates inter-zonal latency and maximizes throughput during active jobs.
The following storage solutions are recommended for optimizing AI and ML system performance with AI zones:
A fully managed, SSD-backed zonal read cache that brings frequently read data from a bucket into the AI zone.
Create an Anywhere Cache instance in an AI zone for the regional source bucket that contains the training datasets or models that you want to serve. When your training job reads a file, the file is pulled into the fast, in-zone cache. Subsequent reads are served directly from the cache, bypassing the regional network. This is ideal for the repetitive data access patterns in model training and for low-latency model serving.
Recommended for:
- Read-heavy workloads
- Low-latency model training and serving
Not recommended for:
- Applications that require full POSIX compliance
Best practices
Follow these best practices for storage when using AI zones:
-
Provision your performance layer in the same AI zone as your compute resources. Colocating compute and storage helps to ensure that GPUs and TPUs remain fully saturated, maximizing "goodput" (useful throughput).
-
For Anywhere Cache, before you start the primary training epoch, perform a pre-read of your dataset to populate, or warm, the SSD-backed cache.
Available AI zones
The following table shows the AI zones and their parent Google Cloud regions.
| Geographic area | Parent region | AI zone |
|---|---|---|
|
United States
|
us-south1
|
us-south1-ai1b
|
Considerations
-
You can access Google Cloud products in a Google Cloud region from the region's AI zone. However, accessing services in a Google Cloud region from an AI zone can add network latency, because the location of the AI zone might be physically separate from the locations of the region's standard zones.
-
We recommend that you run non-ML workloads in standard zones, not AI zones, because AI zones don't offer all Google Cloud services locally.
What's next
- Create a bucket .
- Learn more about Anywhere Cache .
- Read about Cloud Storage bucket locations .
- Read the Architecture Center design guidance for AI and ML workloads .
