forked from graphhopper/graphhopper
-
Notifications
You must be signed in to change notification settings - Fork 1
Performance
karussell edited this page Nov 2, 2012
·
30 revisions
All queries are performed on unterfranken (=> path mean length is ~150km) with a 32bit ubuntu and 1.6.0_33 (same results for jdk1.7.0_05).
The mean query time is printed for some selected algorithms and is in ms. See RoutingAlgorithmIntegrationTests.runShortestPathPerf for more information.
| Storage | -Xmx | Dijkstra | A* | bi-Dijkstra | bi-Dijkstra[1] | bi-Dijkstra[3] | bi-A*[4] | bi-A*[1][4] |
|---|---|---|---|---|---|---|---|---|
| GraphStorage[2] | 300m | 250 | 110 | 190 | 67 | 3 | 85 | 27 |
| Neo4J | 1500m | ~1600 | 730 | 1190 | - | - | - | - |
[2] RAMDirectory is used
[3] contraction hierarchy
[4] paths from bidirectional A* are not always optimal. See code.
Tests are run with an old Intel Core 2 Duo CPU P8600 @ 2.40GHz. E.g. as a reference the encryption tool "john --test" gives me for "Traditional DES":
Many salts: 1044K c/s real, 1050K c/s virtual
Only one salt: 960486 c/s real, 960486 c/s virtual