# Solving the 20-Year S3 File System Problem with Hunter Leath

Page: https://stenobird.com/podcast/screaming-in-the-cloud/solving-the-20-year-s3-file-system-problem-with-hunter-leath
Text version: https://stenobird.com/podcast/screaming-in-the-cloud/solving-the-20-year-s3-file-system-problem-with-hunter-leath.md
Podcast: [Screaming in the Cloud](https://stenobird.com/podcast/screaming-in-the-cloud)
Published: 2026-01-20T11:00:00+00:00
Episode link: https://share.transistor.fm/s/87694d04
Audio file: https://dts.podtrac.com/redirect.mp3/media.transistor.fm/87694d04/81e22e5b.mp3
Processing state: processed
JSON: https://stenobird.com/v1/public/podcasts/screaming-in-the-cloud/episodes/solving-the-20-year-s3-file-system-problem-with-hunter-leath
Duration seconds: 1903

## Resource

Archil solves the long-standing problem of using S3 as a high-performance file system by implementing a high-speed SSD cache layer. This approach allows legacy applications to use S3 with POSIX compatibility and significantly lower costs than EBS or EFS.

## Highlights
- Main idea: Archil acts as a pull-through cache, using fast SSDs to bridge the performance gap between S3 and local disk semantics
- Practical takeaway: Users can attach Archil directly to existing S3 buckets without migrating data, maintaining S3 as the source of truth
- Economic advantage: When accounting for over-provisioning, Archil can cost approximately 1.95 cents per GB, significantly cheaper than EBS
- Failure mode: Traditional attempts to mount S3 as a file system often fail due to high latency and excessive metadata operations (e.g., shell profile reads)
- Future roadmap: The project is currently built on FUSE to allow rapid iteration but aims to move into the Linux kernel as a stable module

## Topics

Cloud Storage, Amazon S3, File Systems, AWS EFS, Infrastructure Engineering, Data Caching, Cloud Economics, Linux Kernel

## Chapters
- 3:30 — The Vision for Archil: Hunter Leath explains the mission to provide EFS-like experiences without the high cost of Amazon's managed services.
- 5:45 — Why S3 Mounting Fails: A look at the technical hurdles of legacy software making thousands of discrete read operations against high-latency object storage.
- 10:30 — The SSD Cache Layer: How using fast SSDs as a cache layer enables high-performance workloads like video transcoding and AI training.
- 12:40 — Integration with Existing Buckets: Details on how Archil attaches to existing production S3 buckets and synchronizes changes back to the source.
- 15:00 — The Economics of Archil vs EBS: A mathematical comparison showing how Archil's pricing beats EBS by eliminating the need for over-provisioning.
- 19:30 — Competing with AWS Features: Discussion on the competitive landscape and the risk of AWS introducing similar caching features.
- 29:10 — Moving Toward the Linux Kernel: The technical roadmap from FUSE-based implementation to a mainline Linux kernel module.

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/screaming-in-the-cloud/episodes/solving-the-20-year-s3-file-system-problem-with-hunter-leath/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/screaming-in-the-cloud/solving-the-20-year-s3-file-system-problem-with-hunter-leath.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.