Database School Building an S3 Competitor with Tigris CEO Ovais Tariq
Nov 6, 2025
Ovais Tariq, co-founder and CEO of Tigris Data and former Uber engineer, shares insights from his vast experience in building hyperscale storage systems. They delve into the intricacies of creating an S3-compatible global object store, discussing Uber's custom hardware designs and the challenges faced during major outages. Ovais highlights Tigris's unique architecture, its focus on AI-driven workloads, and the advantages of its global metadata strategy. He emphasizes the need for specialized storage solutions and the importance of validating ideas before launching.
AI Snips
Chapters
Transcript
Episode notes
FoundationDB As Metadata Backbone
- Tigris uses FoundationDB for strongly consistent, transactional metadata and horizontal scaling.
- FoundationDB's heavy simulation testing increased Tigris's trust in it for global metadata needs.
Two-Tier Metadata Design
- Tigris splits metadata into thin global pointers and richer regional metadata that stores checksums and disk locations.
- Global metadata routes requests to the region that owns the actual bytes for correct locality.
Cache First, Rehome Per-Object
- On a remote read, Tigris serves data transparently and caches frequently accessed objects locally.
- If access persists, Tigris will rehome only the hot object to the new region rather than the entire bucket.