Attribute the partner-side ceiling on cross-site replication throughput

Characterize where on the partner side the MinIO multi-site replication throughput ceiling is established, distinguishing the effects of partner-side hardware, object-store implementation, network-interface capacity, storage-tier I/O operations, replication-state-machine queueing and retry behavior, and the measurement script’s polling interval.

Background

The paper’s remote benchmarks characterize replication as observed at the AIR Data Centre end of transfers from geographically distributed partners. Although the results establish that replication performance varies with the network class traversed by the path, they do not isolate which component on the partner side determines the achievable ceiling.

The authors identify several possible contributors, including partner-side hardware and storage performance, the object-store implementation, NIC capacity, internal queueing and retry behavior in the replication state machine, and the discrete polling interval used to detect destination visibility. A systematic attribution is explicitly deferred to a companion paper using the same dataset.

References

The descriptive cut here cannot resolve where on the partner side the replication throughput ceiling is set: the §7 measurements characterise only the AIR-DC end of each transfer, not the partner endpoint. Candidate factors include partner-side hardware (object-store implementation, NIC capacity, and storage-tier IOPS at the partner endpoint, none of which are characterised in§7), the replication state machine’s internal queueing and retry behaviour, and the discrete polling interval used by the measurement script. A systematic attribution is reserved for a companion paper that exploits the same dataset in greater depth.

— Design and Empirical Evaluation of a Network-Centric, On-Premises Architecture for Earth Observation Data Access  (2608.20283 - Pinelo et al., 20 Aug 2026) in Section 8.5, Relevance to Distributed EO Infrastructure