TechPrufer
Get started
Roblox
Roblox logoRoblox · Feature trail

Roblox Federated Caching Service

1 post · Jun 2025 – Mar 2026

What this trail covers

Roblox's federated caching service has evolved to support record-breaking traffic, handling over 30 million concurrent players and enabling experiences with millions of users playing together. This system acts as a shock absorber for high-velocity, ephemeral data critical for maintaining low latency in Roblox's gaming environment, successfully navigating massive traffic surges and thundering herd situations. The service now sustains 1.38 billion QPS, with single logical clusters exceeding 100 million QPS, and the underlying infrastructure is being enhanced to support even greater scale and resilience during unexpected traffic spikes and thundering herd events.

All Roblox trails
  1. How Roblox’s Cache Sustained 1.38B QPS Beyond Redis LimitsThis post details how Roblox architected a federated caching service to sustain 1.38 billion QPS, exceeding traditional Redis node limits. It addresses the challenge of high-velocity, ephemeral data retrieval in a massive-scale multiplayer gaming environment, where 200ms delays can disrupt gameplay. The system acts as a shock absorber, handling peak loads and navigating a 10x traffic surge on its largest cluster.
  2. The Infrastructure Supporting Record-Breaking ExperiencesThis post details ongoing infrastructure enhancements at Roblox to support record-breaking concurrency, specifically highlighting the platform's ability to handle massive traffic spikes and thundering herd situations. It emphasizes the sum of innovations across the company, rather than a single breakthrough, in building resilient infrastructure. The post mentions supporting experiences like 'Grow a Garden' which achieved a Guinness World Record for most concurrently played video game (21.6 million users), and the platform exceeding 30 million concurrent players. It also touches upon the challenges of supporting millions of creator-built experiences with infrastructure that scales during unexpected traffic spikes, including scenarios where over 21 million users join a single experience simultaneously.