GetBatch: Distributed Multi-Object Retrieval for ML Data Loading
Fuente:
arXiv
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866910033602674688 |
|---|---|
| author | Aizman, Alex Gaikwad, Abhishek Żelasko, Piotr |
| author_facet | Aizman, Alex Gaikwad, Abhishek Żelasko, Piotr |
| contents | Machine learning training pipelines consume data in batches. A single training step may require thousands of samples drawn from shards distributed across a storage cluster. Issuing thousands of individual GET requests incurs per-request overhead that often dominates data transfer time. To solve this problem, we introduce GetBatch - a new object store API that elevates batch retrieval to a first-class storage operation, replacing independent GET operations with a single deterministic, fault-tolerant streaming execution. GetBatch achieves up to 15x throughput improvement for small objects and, in a production training workload, reduces P95 batch retrieval latency by 2x and P99 per-object tail latency by 3.7x compared to individual GET requests. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_22434 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | GetBatch: Distributed Multi-Object Retrieval for ML Data Loading Aizman, Alex Gaikwad, Abhishek Żelasko, Piotr Distributed, Parallel, and Cluster Computing Artificial Intelligence Databases Machine Learning Machine learning training pipelines consume data in batches. A single training step may require thousands of samples drawn from shards distributed across a storage cluster. Issuing thousands of individual GET requests incurs per-request overhead that often dominates data transfer time. To solve this problem, we introduce GetBatch - a new object store API that elevates batch retrieval to a first-class storage operation, replacing independent GET operations with a single deterministic, fault-tolerant streaming execution. GetBatch achieves up to 15x throughput improvement for small objects and, in a production training workload, reduces P95 batch retrieval latency by 2x and P99 per-object tail latency by 3.7x compared to individual GET requests. |
| title | GetBatch: Distributed Multi-Object Retrieval for ML Data Loading |
| topic | Distributed, Parallel, and Cluster Computing Artificial Intelligence Databases Machine Learning |
| url | https://arxiv.org/abs/2602.22434 |