On this page

This page describes the benchmarking methodology used to evaluate Tigris against other object storage providers for small object workloads.

## Benchmark Tool

The [Yahoo Cloud Serving Benchmark (YCSB)](https://en.wikipedia.org/wiki/YCSB) was used to evaluate all systems. We [added support](https://github.com/pingcap/go-ycsb/pull/307) for S3-compatible object storage systems (such as Tigris and Cloudflare R2), which was merged shortly after publishing.

## Test Environment

We ran these benchmarks on compute that was not colocated with storage (i.e., not within Tigris or AWS), matching the workload pattern we see with many of our customers: highly distributed compute across neoclouds.

| Component         | Specification                            |
|-------------------|----------------------------------------|
| Instance type     | VM.Standard.A1.Flex (Oracle Cloud)    |
| Region            | us-sanjose-1 (West Coast)              |
| vCPU cores        | 32                                     |
| Memory            | 32 GiB                                 |
| Network bandwidth  | 32 Gbps                               |

## YCSB Configuration

We benchmarked a dataset of **10 million objects**, each **1 KB** in size. The configuration is available in the [tigrisdata-community/ycsb-benchmarks](https://github.com/tigrisdata-community/ycsb-benchmarks) GitHub repo, specifically at [results/10m-1kb/workloads3](https://github.com/tigrisdata-community/ycsb-benchmarks/blob/main/results/10m-1kb/workloads3).

## Bucket Regions

Buckets were placed in the following regions per provider to ensure fair comparison with geographically proximate endpoints:

| Provider          | Region                                                |
|-------------------|-----------------------------------------------------|
| Tigris            | `auto` (globally replicated, but operating against the `sjc` region) |
| AWS S3           | `us-west-1` (Northern California)                   |
| Cloudflare R2    | `WNAM` (Western North America)                      |

## Test Phases

Using YCSB, we evaluated two distinct phases:

### Phase 1: Bulk Load

Loading **10 million 1 KB objects** into each storage system. This tests raw write performance and system scalability under sustained load.

### Phase 2: Mixed Workload

A mixed workload of **1 million operations** composed of:

- **80% reads** - Simulating typical read-heavy access patterns
- **20% writes** - Representing ongoing data updates

This phase measures real-world performance under typical application workloads.

## Metrics Collected

For each test phase, we collected:

| Metric            | Description                                      |
|-------------------|-------------------------------------------------|
| P50 Latency       | Median latency (50th percentile)                |
| P90 Latency       | Tail latency (90th percentile)                  |
| Runtime           | Total time to complete all operations            |
| Throughput        | Operations per second sustained during the test  |

## Next Steps

View the detailed comparison results for each provider:

- [Comparison: AWS S3](/content/docs/overview/benchmarks/aws-s3/index.html)
- [Comparison: Cloudflare R2](/content/docs/overview/benchmarks/cloudflare-r2/index.html)
- [Model Training on Tigris](/content/docs/overview/benchmarks/model-training/index.html) — ML training workload with TAG caching

Or jump to the [Benchmark Summary](/content/docs/overview/benchmarks/summary/index.html) for complete results.