[mkgmap-dev] new splitter branch solve-parallel

Thu Jul 1 09:50:02 BST 2021

Hi Felix,

yes, sounds reasonable to wait a while if result is not very good. In some cases both algos don't find a good solution, e.g. norway with polygon-file, search-limit=1000000 and --max-nodes=1200000 seems to be stressing, it results in 210 tiles. With search-limit 10 mio it finds a good split with 135 tiles.

Maybe I should simply use a much higher initial search-limit and implement an option to set the maximum time for the search.

Gerd

________________________________________
Von: mkgmap-dev <mkgmap-dev-bounces at lists.mkgmap.org.uk> im Auftrag von Felix Hartmann <extremecarver at gmail.com>
Gesendet: Mittwoch, 30. Juni 2021 21:21
An: Development list for mkgmap
Betreff: Re: [mkgmap-dev] new splitter branch solve-parallel

I feel of all tiles are 85 percent of max then stop. If 5percent of tiles are less the 75 percent max then maybe stop after 30 seconds. If more then 10 percent AND more then 10 tiles are less then 75 percent continue.

On Wed, 30 Jun 2021, 19:09 Gerd Petermann <GPetermann_muenchen at hotmail.com<mailto:GPetermann_muenchen at hotmail.com>> wrote:
Hi all,

I've started this branch to further improve the split results.
Splitter has two different algorithms to find good splits.
1) Algo FULL tries first to split in the middle and then continues with the next positions (mid+1, mid-1, mid+2, mid-2, ...)
The resulting parts are split again recursively. This should find the best possible split but can take very long when the only good split starts far from the middle.

2) Algo SOME uses some heuristics to test only some of the possible splits. This is typically much faster but sometimes doesn't find any solution and sometimes the FULL algo finds a better solution.

The trunk version tries first the one that is more promising and - if that fails to find a good split - tries the other. So, sometimes the FULL algo works for 2 minutes and then SOME finds a good result in a few seconds.

This branch executes both algorithms in parallel and uses the better result.

One of the open problems is to decide under what conditions the execution should stop.
If the SOME algo finds a good solution within 20 secs should we still continue the FULL algo to find the best solution? If yes, how long?
I'm playing with this, any input is welcome.

Gerd
_______________________________________________
mkgmap-dev mailing list
mkgmap-dev at lists.mkgmap.org.uk<mailto:mkgmap-dev at lists.mkgmap.org.uk>
https://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev