# E175: How Dragonfly Is Taking on Redis With a New Data Store

Page: https://stenobird.com/podcast/open-source-startup-podcast/e175-how-dragonfly-is-taking-on-redis-with-a-new-data-store
Text version: https://stenobird.com/podcast/open-source-startup-podcast/e175-how-dragonfly-is-taking-on-redis-with-a-new-data-store.md
Podcast: [Open Source Startup Podcast](https://stenobird.com/podcast/open-source-startup-podcast)
Published: 2025-05-19T20:38:35+00:00
Episode link: https://podcasters.spotify.com/pod/show/ossstartuppodcast/episodes/E175-How-Dragonfly-Is-Taking-on-Redis-With-a-New-Data-Store-e332j0j
Audio file: https://anchor.fm/s/3eab794c/podcast/play/102894035/https%3A%2F%2Fd3ctxlq1ktw2nl.cloudfront.net%2Fstaging%2F2025-4-19%2F7787b13e-391b-5f9f-86ed-bfbbf859597d.mp3
Processing state: processed
JSON: https://stenobird.com/v1/public/podcasts/open-source-startup-podcast/episodes/e175-how-dragonfly-is-taking-on-redis-with-a-new-data-store
Duration seconds: 2382

## Resource

Dragonfly is a multi-threaded, drop-in replacement for Redis designed to handle high-throughput, low-latency workloads. Co-founder Roman Gershman explains how a shared-nothing architecture overcomes the single-threaded bottlenecks that cause instability in traditional in-memory stores.

## Highlights
- Main idea: Redis's single-threaded design creates performance bottlenecks and instability during heavy operations like TTL expiration
- Technical breakthrough: Dragonfly utilizes a shared-nothing architecture where data is partitioned across multiple threads to enable true vertical scaling
- Practical takeaway: A drop-in replacement strategy lowers the barrier to entry for users migrating from established databases like Redis
- Failure mode: Relying on a single main thread for data operations can lead to memory pressure and OOM errors even when total data volume is low
- Business lesson: Building deep infrastructure technology requires a different priority set than building high-margin application layers

## Topics

In-memory data stores, Redis alternatives, Multi-threaded architecture, Open source business models, Database scalability, Infrastructure engineering, Cloud computing, High-performance computing

## Chapters
- 1:00 — The Origin of the Problem: Roman describes his experience at an ad-tech startup where Redis's limitations in high-throughput environments first became apparent.
- 7:15 — The Bottleneck of Single-Threading: An analysis of why Redis's single-threaded architecture fails to scale on modern multi-core hardware, specifically regarding TTL management.
- 10:05 — Viral Growth on GitHub: How Dragonfly gained massive traction and thousands of stars through the developer community and Hacker News.
- 13:05 — The Drop-in Replacement Strategy: Discussing the importance of compatibility in driving early adoption among users of established databases.
- 15:50 — Case Study: Massive Cost Reductions: How a single Dragonfly node replaced a 128-shard Redis cluster, significantly reducing infrastructure complexity and cost.
- 18:40 — Implementing Shared-Nothing Architecture: The technical challenge of building a transactional framework that supports all Redis commands across multiple threads.
- 36:30 — Infrastructure vs. Application Layers: Reflections on the different business priorities required for building deep technology versus high-margin AI applications.

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/open-source-startup-podcast/episodes/e175-how-dragonfly-is-taking-on-redis-with-a-new-data-store/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/open-source-startup-podcast/e175-how-dragonfly-is-taking-on-redis-with-a-new-data-store.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.