Benchmarks | Tigris Object Storage Documentation
On this page
Tigris is highly optimized for storing and retrieving small objects, offering significant performance advantages over other popular object storage solutions.
Performance Highlights
Benchmarks comparing Tigris with AWS S3 and Cloudflare R2 demonstrate that Tigris consistently delivers higher throughput and lower latency for small object workloads:
- Sub-10ms read latency
- Sub-20ms write latency
- 4x throughput compared to AWS S3 for both read and write operations
- 20x throughput compared to Cloudflare R2 for both read and write operations
Why Tigris is Fast
Tigris employs several key strategies to accelerate small object workloads:
- Inlining Small Objects: Very small objects are inlined directly within their metadata records, which minimizes the I/O operations required for retrieval.
- Key Coalescing: Adjacent keys are coalesced to reduce storage fragmentation and overhead, improving data density and scan efficiency.
- On-Disk Caching: Frequently accessed ("hot") items are cached in a high-performance, on-disk cache backed by a Log-Structured Merge-tree (LSM), ensuring rapid access to popular data.
This efficiency allows you to use a single, unified object store for a wide range of object sizes, from very small payloads to large multi-gigabyte blobs, without performance trade-offs.
Benchmark Sections
| Section | Description |
|---|---|
| Performance Metrics | Detailed methodology and benchmark setup |
| Comparison: AWS S3 | Head-to-head performance comparison with AWS S3 |
| Comparison: Cloudflare R2 | Head-to-head performance comparison with R2 |
| Model Training on Tigris | ML training with TAG caching proxy on Tigris |
| Benchmark Summary | Complete results and conclusions |
Reproducibility
You can run these benchmarks yourself using our ycsb-benchmarks repository, which contains the full configuration and instructions.