ml engineering

Storage for training pipelines

DataLoader throughput, checkpoint write patterns, dataset sharing across runs, and the egress costs that show up once your pipeline spans regions.

8 posts

RSS Feed
Friday, June 5, 2026
Training Pipes Team

Kubernetes Persistent Volumes for ML: A Storage Pattern Guide

EBS, EFS, FSx, object storage, CSI drivers — Kubernetes gives you many options for ML storage and all the wrong defaults. Here's the pattern that actually works for training workloads.

Monday, June 1, 2026
Training Pipes Team

Sharing Datasets Across Training Runs Without Copying Terabytes

When five engineers each copy the same 20TB dataset into ephemeral storage, you've got a problem. Here's how to share datasets efficiently across teams and runs.

Thursday, May 28, 2026
Training Pipes Team

The Hidden Cost of Cross-Region Data Egress in ML Pipelines

You don't notice egress until you see the bill. Here's how ML training pipelines quietly rack up cross-region transfer costs, and the architecture that fixes it.

Sunday, May 24, 2026
Training Pipes Team

Checkpointing Large Models: A Storage Guide for ML Engineers

Writing a 500GB checkpoint every hour stresses your storage in ways that training data doesn't. Here's how to design a checkpoint pipeline that's fast, reliable, and doesn't cost a fortune.

Wednesday, May 20, 2026
Training Pipes Team

PyTorch DataLoader Storage Benchmarks: Throughput That Actually Matters

Synthetic storage benchmarks lie about what DataLoader performance feels like in practice. Here's how to measure what your training pipeline actually cares about.

Tuesday, May 12, 2026
Training Pipes Team

How Regional Caching Gateways Cut ML Data Loading Time by 10x

A caching gateway colocated with your GPUs is the biggest single lever for training throughput. Here's how the architecture works and why it produces such dramatic speedups.

Sunday, April 26, 2026
Training Pipes Team

A Practical Guide to Mounting Cloud Storage for GPU Training

Step-by-step guide to mounting cloud storage as a filesystem on your GPU nodes, without the usual FUSE pain or EFS sticker shock.

Sunday, April 19, 2026
Training Pipes Team

Why Object Storage Isn't Enough for Modern ML Training

Object storage like S3 is cheap, durable, and infinite — but it isn't a filesystem. Here's why ML teams keep hitting the same wall, and what to do about it.