# Object Storage Bandwidth Benchmark Baseline

This document captures the reproducible local baseline for performance plan item `Object-storage bandwidth tests`.

Methodology:

- Benchmark exercises the `ObjectStoragePort` transfer paths directly.
- Source-object materialization downloads one `64 MiB` object into worker input storage.
- Generated clip upload writes four clips totaling `32 MiB` back into object storage.
- The baseline uses a local file-copy adapter for raw disk throughput and a throttling layer to simulate object-storage caps without requiring a live S3 environment.

How to rerun:

```bash
mvn -q -Dtest=ObjectStorageBandwidthBenchmarkHarness test
```

Generated artifact:

- `target/benchmarks/object-storage-bandwidth-benchmark.md`

Interpretation notes:

- The `local-disk` row shows the upper bound of the benchmark on the current machine.
- The throttled rows approximate how worker staging and clip upload latency degrade as available storage bandwidth drops.
- This benchmark is intended as a reproducible local planning baseline, not a production throughput claim.

Measured baseline on this workspace:

| Profile | Source MiB | Uploaded MiB | Materialize ms | Upload ms | Materialize MiB/s | Upload MiB/s |
| --- | ---: | ---: | ---: | ---: | ---: | ---: |
| `local-disk` | 64 | 32 | 49.91 | 38.78 | 1282.23 | 825.19 |
| `100 MiB/s` | 64 | 32 | 681.70 | 359.21 | 93.88 | 89.08 |
| `50 MiB/s` | 64 | 32 | 1323.63 | 678.67 | 48.35 | 47.15 |
| `25 MiB/s` | 64 | 32 | 2610.05 | 1322.83 | 24.52 | 24.19 |