S3 over RDMA in SeaweedFS Enterprise is ~5× faster than HTTP for large-object GETs, reaching 11.4 GB/s on a single 100 GbE link.
S3 auth stays intact; data moves directly from storage servers to client memory via RDMA.
SeaweedFS
@SeaweedFSSeaweedFS is a scalable distributed file system optimized for exabyte small files, supports FUSE, S3 API, Hadoop, k8s, replication, Cloud Drive, S3 gateway.
- One table. Multiple query engines. SeaweedFS Table Buckets expose standard Iceberg REST + S3, so the query engine is your choice. Comparing: * Trino for distributed SQL + writes * DuckDB for embedded analytics seaweedfs.com/blog/databow-t… #SeaweedFS #ApacheIceberg #DuckDB #Trino
- Move from MinIO to SeaweedFS? It is just very simple and "works beautifully".
- SeaweedFS Table Buckets can tier each part independently, automatically based on data temperature. ⚡ Metadata → replicated SSD 📦 Parquet data → HDD 🧊 Cold data → erasure coding / cloud ⚡ Parquet footers → cached back on SSD
- Why choose between Iceberg and Lance? You can analyze with Iceberg, and serve with Lance point lookups and vector search, with DuckDB joins them together. SeaweedFS can host both table formats on the same object store, with built-in catalogs for each.
