Historical GeoBench snapshot
Historical evidence—not a current release result. Every figure on this page comes from one Honua-run snapshot dated April 28, 2026. The exact Honua image and full result directory are not publicly pinned, so an independent evaluator cannot reproduce this snapshot exactly. GeoBench also has a directory labeled v2026.1-rc.0 ↗, but its metadata names the mutable nightly-aot image rather than an image digest or immutable public server artifact. Treat it as a labeled snapshot, not a reproducible release-version run.
honua-io/geobench is an Apache-2.0 benchmark harness maintained by Honua for geospatial feature and map servers. Its workloads and configuration are public. The figures below reproduce its published headline snapshot and should be read only within the disclosed test profile and reproducibility limits.
Published source: the GeoBench README's “Current Headline Snapshot” table. It references a generated run directory, results/20260428-192053/, that is not tracked in the public repository. The README therefore preserves the published figures, but not every artifact needed to audit or reproduce the run.
Snapshot conditions (disclosed)
This snapshot was generated on April 28, 2026 HST (April 29 UTC in the report timestamp) against the 100K-point dataset.
- 5-run median per scenario.
- Baseline profile with no spatial response cache.
- 30-second warmup + 30-second measured window per scenario.
- Strict bounded database profile for both servers: Honua active-query / pool settings
6/6/3; GeoServer datastore pool settings6/3. - GeoServer was run with the GSR community extension for the GeoServices rows.
- Server containers, PostGIS, and k6 each use 4 CPU cores and 4 GB memory; each server runs against its own dedicated PostGIS initialized from the same deterministic dataset (seed=42, 100,000 point features, 10 attribute fields).
- QGIS Server remains runnable in the harness, but the headline table focuses on the Honua vs. GeoServer profile.
Servers in the harness: Honua Server (.NET 10), GeoServer 2.28.0 (Java/JVM), QGIS Server 3.38 (C++/Qt).
What this snapshot reported
Within this April 28 test profile, the published table reports that Honua led every comparable performance cell against GeoServer; the six comparable error-rate cells were tied at 0.0%.
MapServer/export has no GeoServer row in this harness profile. These results describe this run—not general performance across versions, data models, infrastructure, or production workloads.
Advantage by scenario — April 28, 2026 snapshot
| Scenario | Throughput Advantage | Tail-Latency Advantage |
|---|---|---|
GeoServices MapServer/identify, large bbox | 470x higher req/s | 450x lower p95 |
WMS filtered GetMap, range | 101x higher req/s | 113x lower p95 |
| Attribute filter, LIKE | 73x higher req/s | 185x lower p95 |
| Spatial bbox, large bbox | 55x higher req/s | 63x lower p95 |
| WFS filtered, LIKE | 20x higher req/s | 20x lower p95 |
| WMS reprojection, large bbox | 14x higher req/s | 24x lower p95 |
| Concurrent mixed workload, 100 VUs | 12x higher req/s | 8.9x lower p99 |
WFS GetFeature, large bbox | 10x higher req/s | 14x lower p95 |
GeoServices FeatureServer/query, medium bbox | 4.1x higher req/s | 5.2x lower p95 |
WMS GetMap, medium bbox | 2.6x higher req/s | 3.1x lower p95 |
Raw snapshot values — April 28, 2026
| Track | Scenario | Honua | GeoServer |
|---|---|---|---|
| Attribute filter | LIKE | 1632.8 req/s, p95 5.2 ms | 22.3 req/s, p95 959.3 ms |
| Spatial bbox | large bbox | 1069.3 req/s, p95 15.5 ms | 19.5 req/s, p95 977.1 ms |
| Concurrent mixed workload | 100 VUs | 1190.2 req/s, p99 201.4 ms | 96.2 req/s, p99 1793.2 ms |
WMS GetMap | medium bbox | 80.1 req/s, p95 186.4 ms | 30.5 req/s, p95 582.5 ms |
| WMS reprojection | large bbox | 131.9 req/s, p95 99.3 ms | 9.3 req/s, p95 2356.7 ms |
WFS GetFeature | large bbox | 581.5 req/s, p95 29.0 ms | 57.8 req/s, p95 418.5 ms |
| WFS filtered | LIKE | 1094.3 req/s, p95 13.2 ms | 54.6 req/s, p95 269.1 ms |
WMS GetFeatureInfo · superseded | medium bbox | Historical: 2808.9 req/s, p95 4.9 ms | Historical: 407.1 req/s, p95 53.7 ms |
WMS filtered GetMap | range | 171.4 req/s, p95 105.5 ms | 1.7 req/s, p95 11953.6 ms |
GeoServices FeatureServer/query | medium bbox | 937.2 req/s, p95 13.8 ms | 228.9 req/s, p95 71.7 ms |
GeoServices MapServer/identify | large bbox | 2868.9 req/s, p95 6.3 ms | 6.1 req/s, p95 2833.6 ms |
GeoServices MapServer/export | large bbox | 152.2 req/s, p95 108.6 ms | No comparison row in this snapshot |
Superseded row: the historical headline table included wms-getfeatureinfo, but the current matrix marks that scenario blocked and non-comparable. It is retained only as part of the historical raw record and is excluded from the advantage summary.
Limits to apply before using these results
- Response-shape audits are included in the report. Some feature and native rows show payload metadata or property-key drift, so public claims should include those caveats.
- Exact reproduction is not currently possible from public artifacts. The Honua image,
honua-geobench:trunk-b650a321-rendergate2, was a local build from source aroundb650a321. A matching source revision and immutable image must be public before this snapshot can be independently re-run exactly. - The GeoServer image resolved locally to digest
sha256:48fcd9488f35c29ef8b8dd2d0b6ae491d1bef73cea83f0ef27f6fa124ddcf245, run withGEOSERVER_COMMUNITY_EXTENSIONS=gsrfor the GeoServices rows. - The GeoBench matrix status ↗ marks
wms-getfeatureinfoblocked on Honua in a later rerun context (HTTP 405 in one harness profile). That status supersedes the historical value above for comparison purposes.
Run the current harness
GeoBench is Apache-2.0 and can be run against available images. This exercises the current harness; it does not reproduce the historical snapshot exactly without the missing pinned Honua image and generated result artifacts.
git clone https://github.com/honua-io/geobench
cd geobench
python3 data/small/generate.py
./scripts/run-benchmark.shThe two-server profile, pool settings, warmup, and duration are documented in the GeoBench README. Check the image references before comparing a new run with the historical figures above.
Cost and price
Performance numbers only matter against what they cost to run. Honua's commercial position is a server that is source-available and self-hostable under the Elastic License 2.0 (ELv2) ↗, with Apache-2.0 SDKs, which changes the cost basis versus a per-named-user proprietary stack:
- License boundary: an ELv2 server you can inspect and self-host, with Apache-2.0 SDKs; paid entitlements are validated locally without a connectivity-dependent license service.
- No per-named-user license fee for client access. ArcGIS Pro, the ArcGIS API for Python, QGIS, and OGC/STAC clients connect to the same server without per-seat client entitlements on Honua's side.
- In the April 28 snapshot, the attribute-filter LIKE scenario reported 1632.8 req/s for Honua and 22.3 req/s for GeoServer under the same container resource limits and dataset, with the disclosed
6/6/3and6/3pool settings. That result is specific to this historical profile.
For current capacity bands, editions, and worked licensing examples, see the public pricing page.
No TCO or cost-normalized result is claimed. GeoBench does not publish an apples-to-apples competitor price model, requests per provisioned vCPU, or equivalent cloud-cost metric.