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.
@article{arxiv.2602.22434,
title = {GetBatch: Distributed Multi-Object Retrieval for ML Data Loading},
author = {Alex Aizman and Abhishek Gaikwad and Piotr Żelasko},
journal= {arXiv preprint arXiv:2602.22434},
year = {2026}
}