Performance |
Top Previous Next |
Performance of the core routing functionality is important to anyone considering using large street networks.
From the TIGER data (USA) we have extracted a 4,500,000 links street network covering the central part of USA (1600 x 1600 km).
The test was done with this setup:
•Dynamic segmentation, 2 calls to function Coordinate2location •Generation of driving directions, function RouteList •Latitude/Longitude coordinates •Alpha = 1.3 •Fastest route •Caching of all data, except file coord3.bin •Computer was an AMD Athlon 64 3200+ with 2 GB RAM. •The network required 450 MB harddisk space.
We have randomly calculated a large number of routes.
Average time per route:
Long routes can be as long as 3100 km, but most will be a lot shorter. Short routes are within the same network, but start and end location is restricted to a 500 x 500 km rectangle with 450,000 links.
Multi-threading
The normal-mode, long route test from above has been executed on a number of different computers and with different number of threads. As can be seen from the table below, true dual CPU gives more processing power than a hyper-threading enabled (HT) CPU. Of course this requires simoultaneous routing requests and an application allowing for it, such as RW NetServer:
Number of routes per minute:
Local vs. remote connection
Timings above are for local connections (server address = 127.0.0.1 / localhost). If you are making a remote connection across the internet or LAN, each function call may take up to 1 second extra, depending on the exact setup and network configuration. |