Taming Database Latency: When External Caching Backfires (And How to Fix It)

Milliseconds matter. Long-tail latency (p99) can dictate your user’s experience and bottom line. When database queries slow down, a common practice is to throw an external cache like Redis in front. But what happens when that external cache introduces more complexity, cache-invalidation nightmares, and hidden costs?
In this talk, Guy Shtub explores database latency mechanics, common caching patterns, and why adding an external cache tier doesn't always solve your performance bottlenecks.

### What you'll learn:

• The Real Business Cost of Latency: How tail-latency (p99) impacts revenue and why the slowest 1% dominates user experience.
• Caching Strategies in the Wild: Breaking down Cache-Aside, Write-Through, Write-Back, and Embedded Read-Through patterns.
• The Dark Side of External Caching: How external cache tiers add application complexity, degrade availability, and ignore rich database context.
• Bypassing the Cache Tier: How modern, close-to-metal database designs (shard-per-core, unified cache, I/O scheduling) enable databases to replace external cache tiers altogether.
• Real-World Case Studies: How teams at Comcast, Disney+/Hotstar, Grab, and Tubi eliminated external cache layers to boost performance and cut infrastructure TCO.

Speaker: Guy Shtub, Head of Training at ScyllaDB
Guy Shtub is Head of Training at ScyllaDB and holds a B.SC. degree in Software Engineering from Ben Gurion University. He co-founded two start-ups and is experienced in creating products that people love.