forked from graphhopper/graphhopper
-
Notifications
You must be signed in to change notification settings - Fork 1
Performance
karussell edited this page Aug 30, 2012
·
30 revisions
All queries are performed on unterfranken (=> path mean length is ~150km) with linux and 1.6.0_33 (same results for jdk1.7.0_05). Before measuring query time warmup queries (50) are performed. then 100 random queries are performed, but for every storage the same.
The mean query time is printed for some selected algorithms and is in seconds.
Storage | -Xmx | dijkstra | astar | bidijkstra | bidijkstra skip** |
---|---|---|---|---|---|
MemoryGraphSafe | 300m | 0.45 | 0.10 | 0.13 | 0.05 |
MMapGraph | 300m | 0.51 | 0.12 | 0.19 | nyd* |
Neo4J | 1500m | ~1.6 | 0.73 | 1.19 | nyd* |
- nyd = not yet done ** bi dijkstra 2-skip = bidirectional dijkstra with shortcuts for every 2-degree node
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