[
https://issues.apache.org/jira/browse/LUCENE-7195?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15233476#comment-15233476
]
Michael McCandless commented on LUCENE-7195:
--------------------------------------------
Thanks for fixing this [~daddywri], it let me run the geo bench again!
Our geobench (in luceneutil) now has so many options that I wrote a wrapper
Python script (also pushed to luceneutil) to run it on the N possibilities and
summarizes the results
These results are on the 60.8 M point London, UK extract (100% of London, 2.5%
of the rest of the world) from the OpenStreetMaps export:
h3. Indexing performance:
||Approach||Index time (sec)||Force merge time (sec)||Index size (GB)||Reader
heap (MB)||
|Geo3d|577.8s|281.1s|0.80|1.96|
|Points|354.8s|165.0s|1.05|1.96|
|Geopoint|119.0s|15.8s|0.75|2.73|
Here {{geopoint}} wins big since it builds on postings which are much faster at
merging than BKD trees, but note that the benchmark intentionally picks
settings that provoke heavy merging (small indexing buffer, serial merge
scheduler). It also shows that using the same morton code, but on (1D) points
instead of postings, should be a really good option too.
h3. Search performance, if queries are NOT pre-built (i.e., query construction
time counts):
||Shape||Approach||QPS||Hit count||
|Distance|Geo3d|29.65|383371933|
|Distance|Points|28.65|382961957|
|Distance|Geopoint|17.88|382961953|
|Box|Geo3d|45.31|221118844|
|Box|Points|68.11|221118844|
|Box|Geopoint|38.09|221118851|
|Polygon (10 sides)|Geo3d|5.94|355855224|
|Polygon (10 sides)|Points|29.26|355809475|
|Polygon (10 sides)|Geopoint|19.43|355809465|
h3. Search performance if queries are pre-built:
||Shape||Approach||QPS||Hit count||
|Distance|Geo3d|31.89|383371933|
|Distance|Points|28.84|382961957|
|Distance|Geopoint|17.50|382961953|
|Box|Geo3d|45.07|221118844|
|Box|Points|69.78|221118844|
|Box|Geopoint|35.75|221118851|
|Polygon (10 sides)|Geo3d|24.88|355855224|
|Polygon (10 sides)|Points|29.12|355809475|
|Polygon (10 sides)|Geopoint|19.29|355809465|
Pre-building makes a huge difference in {{Geo3d}} for polygons, and curiously
seems not to matter much for {{Points}} and {{Geopoint}}.
> GeoPolygon construction sometimes inexplicably chooses concave polygons when
> order of points is clockwise
> ---------------------------------------------------------------------------------------------------------
>
> Key: LUCENE-7195
> URL: https://issues.apache.org/jira/browse/LUCENE-7195
> Project: Lucene - Core
> Issue Type: Bug
> Components: modules/spatial3d
> Affects Versions: master
> Reporter: Karl Wright
> Assignee: Karl Wright
> Fix For: master, 6.x
>
> Attachments: LUCENE-7195.patch
>
>
> The following input generates the following polygon, which is backwards from
> the correct sense:
> {code}
> MAKE POLY: centerLat=51.204382859999996 centerLon=0.23125274999999992
> radiusMeters=44832.90297079173 gons=10
> polyLats=[51.204382859999996, 50.89947531437482,
> 50.8093624806861,50.8093624806861, 50.89947531437482, 51.204382859999996,
> 51.51015366140113, 51.59953838204167, 51.59953838204167, 51.51015366140113,
> 51.204382859999996]
> polyLons=[0.8747711978759765, 0.6509219832137298, 0.35960265165247807,
> 0.10290284834752167, -0.18841648321373008, -0.41226569787597667,
> -0.18960465285650027, 0.10285893781346236, 0.35964656218653757,
> 0.6521101528565002, 0.8747711978759765]
> --> QUERY: PointInGeo3DShapeQuery: field=point: Shape:
> GeoCompositeMembershipShape: {[GeoConcavePolygon:
> {planetmodel=PlanetModel.WGS84, points=
> [[lat=0.899021779599662, lon=0.011381469253029434],
> [lat=0.9005818372758149, lon=0.006277016653633617],
> [lat=0.9005818372758149, lon=0.0017952271299490152],
> [lat=0.899021779599662, lon=-0.003309225469446801],
> [lat=0.8936850723587506, lon=-0.007195393820967987],
> [lat=0.8883634317734164, lon=-0.0032884879971082164],
> [lat=0.8867906661272461, lon=0.0017959935133446592],
> [lat=0.8867906661272461, lon=0.006276250270237971],
> [lat=0.8883634317734164, lon=0.011360731780690846],
> [lat=0.8936850723587506, lon=0.015267637604550617]], internalEdges={},
> holes=[]}]}
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
