OSM Network
OSM load speed: server vs remote client.
The network study uses an R9 9950X client loading onto an i5-13600K server while throttling the link from 100Mb/s through 10Gb/s. Dotted reference lines show the same server with a host-local client (646 kNodes/s nodes, 274 kNodes/s index). Even at 10Gb/s, index build on the remote client (208 kNodes/s) stays below that i5 host-local baseline; node loading reaches 666 kNodes/s, in line with the on-server client. Below 2.5Gb/s, both phases degrade—the 100Mb/s case drops to 90 kNodes/s.
Comparing host-local results across different CPUs (for example R9 vs i5) is not apples-to-apples here; a fair cross-system claim would need matched client/server hardware or bidirectional remote tests in both directions.
Regenerate: python3 reports/osm-network-speed.py (reads docs/results-summary/pg18/osm-network.md).
Snapshot of the Streamlit explorer — down between releases (OSM Network) on 2026-08-09. Captured by explorer/snapshot-streamlit.py from the live site.
Query returned 9 rows.
| client | server | conn | nodes_kips | index_kips | wal_mbps |
|---|---|---|---|---|---|
| Apple M4 Max 128GB Apple AP2048Z | Apple M4 Max | host | 463.0 | 273.0 | 12.8 |
| R9 9950X 128GB SK51 2TB | Apple M4 Max Studio | 10000 | 614.0 | 303.0 | 16.9 |
| R9 9950X 124GB SK51 2TB | R9 9950X | host | 758.0 | 366.0 | 21.2 |
| R9 9950X 128GB SK51 2TB | i5-13600K | 10000 | 666.0 | 208.0 | 18.4 |
| R9 9950X 128GB SK51 2TB | i5-13600K | 5000 | 625.0 | 245.0 | 17.3 |
| R9 9950X 128GB SK51 2TB | i5-13600K | 2500 | 505.0 | 204.0 | 14.7 |
| R9 9950X 128GB SK51 2TB | i5-13600K | 1000 | 428.0 | 204.0 | 12.4 |
| R9 9950X 128GB SK51 2TB | i5-13600K | 100 | 90.0 | 122.0 | 2.6 |
| i5-13600K 188GB Fury G5 4TB | i5-13600K | host | 646.0 | 274.0 | 17.8 |