On 6/1/12 1:21 PM, Stephen Woodbridge wrote:
On 6/1/2012 11:45 AM, Ian wrote:
Hi all,
Does anyone have any estimations on how long it would take to run
assign_vertex_id on the entire Census TIGER dataset for the United States?
Say you had a 2.6ghz intel core 2 duo processor and 8gb of RAM? Or has
anybody done this and have a time and their computer's specs?
I'm going to take a wild guess that it is something in the ball park of 1 week. You have a secondary problem in that this is all done in a single stored procedure call, which means that it is all done in a single transaction. This might be an issue as the transaction will become extremely large.
You might look at using the TNIDF and TNIDT which are the Tiger assign node numbers.
So how are you planning on using this data with pgRouting? Tiger does not have attributes like oneway streets, zlevels at intersections or turn restrictions all of which are needed to do useful routing planning.
Are you aware the using tiger you will generate routes like enter an exit of highway, drive to the top of an overpas and make a left hand turn onto the underpass 20 feet below you, etc?
It can be useful for some restricted routing uses where you can overlook these issues but in general it is not suitable for most routing tasks.
-Steve
_______________________________________________
Pgrouting-users mailing list
Pgrouting-users@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/pgrouting-users
Thanks for the input.
So TNIDF & TNIDT can be imported to the 'source' & 'target' columns as long as they are properly linked to the endpoint geometry?
My project has to do with walking in cities so I'm thinking that I will be able to overlook routing issues associated with driving....I will have to make some assumptions though (sidewalks ped-xings), but streets in cities tend to be walkable.
I noticed that OSM data has some attribution TIGER data does not, maybe I should look there too.
Even if I clip the streets to only include the areas I really need (cities with a population of over 10,000), I'm guessing that would still be the majority of all the streets (maybe 1/3 or more?)
Well... if I end up doing assign_vertex_id on a large dataset I'll report back any useful results,
ian