[Routing] [OSM-dev] pre-compute routing

[Jon Bright]

Hi,

Jon Bright wrote:
 >
> I want this one to get finished first).  But I'm going to go out on a 
> limb and say that we have about a million fork nodes.

In fact, it's 3255678 fork nodes (based on a 2-week-old planet).

> Going with the million-fork-node assumption from above, all-pairs would 
> be 5*(1000000*999999)/8 ~= 625,000,000,000 routes.  Assume 75% of these 
> are unroutable (different continents, mixing of cycle paths and 
> motorways, etc.) and you have 156,250,000,000 routes.

5*(3255678*3255677)/8 ~= 6,624,647,490,003 routes.  Assume 75%...

~= 1,656,161,872,500

> By my calculations thus far:
> 
> ~= 1.5625 * 10e11 * (L * 0.5) * N

~= 1.6561 * 10e12 * (L * 0.5) * N

> ~= 1.5625 * 10e11 * (20 * 0.5) * 7
> 
> ~= 10,937,500,000,000 bytes
> 
> ~= 10,430,813 MB
> 
> ~= 10,186 GB
> 
> ~= 10TB

Assuming N as 1 (bearing in mind Robert's mail about many routes being 
dupes) and dropping 20 to 5 based on the same mail:

~= 4,140,404,681,252 bytes

~= 3,948,597 MB

~= 3,856 GB

~= 3TB

I'm still not sure this approach is worthwhile.  I don't think I'll be 
following it up for the moment.

--
Jon