Hi Lambertus, the algo is quite simple. The max-nodes parameter is ignored, splitter calculates an initial max-nodes value by dividing the total number of nodes by the wanted number of tiles. Then the normal split algo is used. If the result has too many tiles, the max-nodes value is increased, if it is too low, it is decreased. This is repeated until either a solution is found or no max-nodes value is found that results in the number of tiles. In other words: the tiles are as "equally sized" as before. Gerd Lambertus wrote
Thanks Gerd, seems to be working great!
Two questions: - How does this interact with the --max-nodes setting? - Could it be used for an initial split of the entire planet? I.e., set the number of tiles I want and have splitter output equally sized tiles? (I know I could just try, but the servers are busy)
On 2014-05-05 09:28, Gerd Petermann wrote:
Hi Lambertus,
attached is a patch that implements the new option: --num-tiles A target value that is used when no split-file is given. Splitting is done so that the given number of tiles is produced. The max-nodes value is ignored if this option is given. A compiled binary is here:
http://files.mkgmap.org.uk/download/206/splitter.jar [4]
Please let me know if it works as expected.
Gerd
------------------------- Date: Sun, 4 May 2014 22:20:20 +0200 From:
osm@
To:
mkgmap-dev@.org
Subject: Re: [mkgmap-dev] mkgmap ToDo list
Great, many thanks. I'm currently trying to create as large as (automatically) possible contour tiles and a lot of time is put in subsplitting a tile again and again until two subtiles appear. This new option would save much time.
On 05/04/2014 10:13 PM, Gerd Petermann wrote:
Hi Lambertus,
yes, I want to code that tomorrow. I prefer to make it "give me n tiles", but if that turns out to be difficult I try the simple version. I can think of scenarios where it is not possible to create exactly n tiles.
Gerd
------------------------- Date: Sun, 4 May 2014 22:08:32 +0200 From:
osm@
To:
mkgmap-dev@.org
Subject: Re: [mkgmap-dev] mkgmap ToDo list
Is the 'just give me two tiles' option for splitter on the to-do list? I'd really appreciate such a function.
On 05/04/2014 09:36 PM, Gerd Petermann wrote:
Hi Felix,
would using the precomp-sea parameter for splitter improve results? It's not that tiles with loads of sea actually failed.
good question. I've coded that parameter in splitter before we did some important changes in mkgmap. I think it is needed when you use a polygon file in combination with an imput file that doesn't cover the polygon. If the polygon covers a costline with a lot of islands, but the input file doesn't contain the data, splitter might create large tiles covering these "empty" areas. Later, in mkgmap, the empty areas are filled with the precompiled-sea data, and that can cause too large files.
BTW: The quick hack turned out to be useless. It worked better in some areas and worse in others. I'll look again at this later next week.
Gerd
------------------------- Date: Sun, 4 May 2014 21:21:59 +0200 From:
extremecarver@
To:
mkgmap-dev@.org
Subject: Re: [mkgmap-dev] mkgmap ToDo list
No - I use splitter like this - with maxnodes depending on area - I usually just reduced it after failures to compile a tile.: java -Xms4000m -Xmx10400m -jar c:openmtbmapsplitter.jar --max-nodes=%maxnodes% --max-threads=8 --output=pbf --keep-complete --max-areas=1524 --geonames-file=cities5000 --description=%country% --mapid=%FID%0000 c:OpenMTBMaposmpbf_geofabrik%country%-latest.osm.pbf would using the precomp-sea parameter for splitter improve results? It's not that tiles with loads of sea actually failed.
On 03.05.2014 09:00, Gerd Petermann wrote:
Hi all,
I've coded a quick hack which seems to improve the ratios. On the other hand, I don't see these large differences between smallest and largest img file. What part of the world should I try? Do you use the precomp-sea parameter in splitter?
Gerd
------------------------- Date: Wed, 30 Apr 2014 14:59:32 +0200 From:
extremecarver@
To:
mkgmap-dev@.org
Subject: Re: [mkgmap-dev] mkgmap ToDo list
if that doesn't seriously (more than 30-40%) slow down the splitter, I assume it would be much better...
On 30.04.2014 14:06, Gerd Petermann wrote:
Hi,
if I got that right the number of nodes is not highly correlated to the img size, so the max-nodes value is not a good estimate.
I assume the reason is that nodes which belong to roads produce a lot more bytes in the img file compared to nodes which are parts of shapes or other non-routable ways, not talking about nodes which are simply ignored by the style.
So, a possible solution in splitter could be to parse the ways before reading the nodes and save all nodeids which belong to ways with highway=*. If these nodes are refered by more than one way with highway=* we assume that they will be routing nodes. These special nodes could be counted e.g. 10 times to give a better estimate.
Gerd
Date: Wed, 30 Apr 2014 13:36:19 +0200 From:
osm@
To:
mkgmap-dev@.org
Subject: Re: [mkgmap-dev] mkgmap ToDo list
Multithreading the tile rendering for a single map is indeed a bit difficult and I gave it up because you need to keep track which image id's are already in use. Since I provide multiple maps the work-around is running a few scripts parallel, which is also a crude form of multithreading.
The script language is PHP and it doesn't run on Windows without some changes ('/' vs '' in paths, 'rm -rf', that sort of stuff). Never tried it.
To get a better optimum in file size, using the process I described earlier, you could start off with a huge --max-nodes setting and then 'search' for the highest --max-nodes that works for each specific area.
On 30/04/2014 11:49, Felix Hartmann wrote:
I would love if there was a possibility that you pass the used max-nodes value to mkgmap. When mkgmap is compiling the maps, then after the .img is created it should check a) did it crash due to too many max-nodes b) for me not important - but for others with very old GPS, etrex 10, ---> is tile bigger than X (usually 8) MB.
if a) or b) true, then pass the file back to splitter and split with 60% of maxnodes - and compile the resulting .img files again. Should it fail again, use 40%, again 25%... Sometimes there are awful tiles, that need supersmall max-nodes till they compile, however lately (last 1-2 years) I never encountered them anymore. I think that happened rather due to a but in splitter/mgkmap that is fixed now.
okay, you could also do this with a script, but it gets rather complicated to multithread it (you need to wait till mgkmap finished compiling all .img files - and run mkgmap first without address index to save time) and do some clever routines on making sure that the map id (e.g. 6340????.img) stay correct. Even more complicated to have consequent map id...
For europe with a fixed max-node I get tiles from 1.9MB up to 18MB. That's factor 9 - so it's huge... If I could narrows that down easily to 8-18MB - without getting tiles crashing due to too high max-nodes values, that would be sweet.
As for the scripts - would they run on windows too? - What programming language are they in?
On 29.04.2014 21:39,
osm@
wrote:
Oh, and ofcourse anyone interested can get my scripts, send an email. They'll be on Github someday anyway.
On 2014-04-29 20:37, Gerd Petermann wrote: > Hi Lambertus, > > okay, if I got that right you finally get *.img files with a size > near (but below) 8MB, so maybe Henning can use that script, too. > > If you do that for e.g. Germany, how small is tpically the smallest > *.img file ? > Is it probably near 4 MB? > > Gerd > >> To:
mkgmap-dev@.org
>> Date: Tue, 29 Apr 2014 20:30:27 +0200 >> From:
osm@
>> Subject: Re: [mkgmap-dev] mkgmap ToDo list >> >> These are the direct results from Splitter. The format is o5m, both >> input as output. Splitter version is: r321. >> >> For this test I split the original source with --max-nodes=8000000. > Then >> I render the initial tiles, when the result is larger than 8MB it's >> subsplit again with --max-nodes=(8000000-(attempt*100000)). The > initial >> source files are ~70MB (o5m) and after several subsplits the two > *.img >> are < 8MB. During this process --max-nodes has been reduced to e.g. >> 7500000 and the source file is split up in two o5m files of about > 37MB. >> >> I can upload an example source file and it's two subsplit siblings > if >> you want. >> >> >> On 2014-04-29 19:38, GerdP wrote: >> > Hi Lambertus, >> > >> > that's interesting. Are these the img file sizes or the osm file > sizes? >> > >> > Gerd >> > >> > >> > Lambertus wrote >> >> Unfortunately I cannot confirm that. Below is a bit of logging > from my >> >> script: >> >> Original: 97000020 (70551453), New: 0 (35684445), New: 1 > (36852845) >> >> Original: 97000001 (74621042), New: 0 (37522992), New: 1 > (37222739) >> >> Original: 97000002 (73391358), New: 0 (37679505), New: 1 > (38098627) >> >> Original: 97000003 (77862567), New: 0 (39075311), New: 1 > (39261197) >> >> >> >> The original files above contain contour data, the filesize is > between >> >> brackets. As you can see both resulting file are approximately > the >> >> same >> >> size. >> >> >> >> On 2014-04-29 15:39, Gerd Petermann wrote: >> >>> Hi Lambertus, >> >>> >> >>> and I guess that even after this optimization you will >> >>> see a factor 3 or higher between the largest tile and the > smallest. >> >>> Can you confirm that? >> >>> >> >>> Gerd >> >>> >> >>>> Date: Tue, 29 Apr 2014 15:32:38 +0200 >> >>>> From: >> > >> >> osm@ >> > >> >>>> To: >> > >> >> mkgmap-dev@.org >> > >> >>>> Subject: Re: [mkgmap-dev] mkgmap ToDo list >> >>>> >> >>>> Num-tiles=x would indeed be better for this specific need. >> >>>> >> >>>> It is my experience that it regularly takes multiple calls to >> >>> Splitter >> >>>> to get 2+ sub-tiles when you reduce the max-nodes by 100k for > each >> >>>> sub-split attempt. This is what I currently do to get an > optimum in >> >>>> tile-size vs total number of tiles. >> >>>> >> >>>> >> >>>> >> >>>> On 29/04/2014 15:09, Gerd Petermann wrote: >> >>>> > Hi Lambertus, >> >>>> > >> >>>> > that sounds like a possible change in splitter: >> >>>> > Instead of specifying max-nodes you may specify --num-tiles=x >> >>>> > and splitter will try to find a split that produces excactly > x >> >>> tiles >> >>>> > which are not too narrow and have a node number which is not >> >>>> > too far from the average (but still aligned to a multiple of > map >> >>> units >> >>>> > as now). >> >>>> > So, for your script that means you don't have to find the >> >>> max-nodes >> >>>> > value. >> >>>> > >> >>>> > I'll think about this again... >> >>>> > >> >>>> > Gerd >> >>>> > >> >>>> > > Date: Tue, 29 Apr 2014 14:59:36 +0200 >> >>>> > > From: >> > >> >> osm@ >> > >> >>>> > > To: >> > >> >> mkgmap-dev@.org >> > >> >>>> > > Subject: Re: [mkgmap-dev] mkgmap ToDo list >> >>>> > > >> >>>> > > While this possibly can be solved in Splitter or Mkgmap, it >> >>> could also >> >>>> > > be solved by your build-script when you add a maximum tile > size >> >>> check >> >>>> > > and re-split (with a lower number of max-nodes) until you > get >> >>> two or >> >>>> > > more sub-tiles. Granted, this adds complexity to the script > but >> >>> it works >> >>>> > > well for me. >> >>>> > > >> >>>> > > On 25/04/2014 21:54, Henning Scholland wrote: >> >>>> > > > Hi Gerd, >> >>>> > > > >> >>>> > > > I would like to have img-tiles which have globally nearly > the >> >>> same >> >>>> > > > filesize, so that they use the space of devices like > eTrex 10. >> >>>> > > > >> >>>> > > > With my actual map I use globally the same value for >> >>> max-nodes. But the >> >>>> > > > size of the img-tiles differ more then factor 2. Eg. a > tile in >> >>> Germany >> >>>> > > > is between 2 and 5 mb where a tile in China is about 10 > mb. If >> >>> I remove >> >>>> > > > details, this difference will increase, because in > Germany >> >>> more objects >> >>>> > > > will be removed from the img-tile then in China. >> >>>> > > > >> >>>> > > > Henning >> >>>> > > > >> >>>> > > > _______________________________________________ >> >>>> > > > mkgmap-dev mailing list >> >>>> > > > >> > >> >> mkgmap-dev@.org >> > >> >>>> > > > http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] >> >>>> > > >> >>>> > > _______________________________________________ >> >>>> > > mkgmap-dev mailing list >> >>>> > > >> > >> >> mkgmap-dev@.org >> > >> >>>> > > http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] >> >>>> > >> >>>> > >> >>>> > _______________________________________________ >> >>>> > mkgmap-dev mailing list >> >>>> > >> > >> >> mkgmap-dev@.org >> > >> >>>> > http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] >> >>>> > >> >>>> >> >>>> _______________________________________________ >> >>>> mkgmap-dev mailing list >> >>>> >> > >> >> mkgmap-dev@.org >> > >> >>>> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] >> >>> >> >>> _______________________________________________ >> >>> mkgmap-dev mailing list >> >>> >> > >> >> mkgmap-dev@.org >> > >> >>> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] >> >> _______________________________________________ >> >> mkgmap-dev mailing list >> > >> >> mkgmap-dev@.org >> > >> >> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] >> > >> > >> > >> > >> > >> > -- >> > View this message in context: >> > >
http://gis.19327.n5.nabble.com/mkgmap-ToDo-list-tp5803388p5804588.html
[3]
>> > Sent from the Mkgmap Development mailing list archive at > Nabble.com. >> > _______________________________________________ >> > mkgmap-dev mailing list >> >
mkgmap-dev@.org
>> > http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] >> _______________________________________________ >> mkgmap-dev mailing list >>
mkgmap-dev@.org
>> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] > > _______________________________________________ > mkgmap-dev mailing list >
mkgmap-dev@.org
> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1] _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1]
-- keep on biking and discovering new trails
Felix openmtbmap.org & www.velomap.org [2]
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1]
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
-- keep on biking and discovering new trails
Felix openmtbmap.org & www.velomap.org [2]
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1]
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1]
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1]
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [1]
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
Links: ------ [1] http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev [2] http://www.velomap.org [3] http://gis.19327.n5.nabble.com/mkgmap-ToDo-list-tp5803388p5804588.html [4] http://files.mkgmap.org.uk/download/206/splitter.jar
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
mkgmap-dev mailing list
mkgmap-dev@.org
-- View this message in context: http://gis.19327.n5.nabble.com/mkgmap-ToDo-list-tp5803388p5805281.html Sent from the Mkgmap Development mailing list archive at Nabble.com.