logo separator

[mkgmap-dev] splitter r325: improved split algo and new option

From Gerd Petermann gpetermann_muenchen at hotmail.com on Wed May 7 15:43:46 BST 2014

Hi Felix,

I seem to remember that the disadvantage of pbf for output of splitter was higher,
but maybe that was before I modified splitter to use a smaller "BatchLimit"
in the pbf write routine:
configBatchLimit(1000);

The default value produced slightly better compression but required much more heap.
The original code was optimized for one write process, while splitter may start many
hundrets, and in some situations splitter and pbf writer were fighting for the heap,
causing heavy work for GC. Of course this only happened when splitting large files
like Europe or planet.

Gerd

Date: Wed, 7 May 2014 16:06:27 +0200
From: extremecarver at gmail.com
To: mkgmap-dev at lists.mkgmap.org.uk
Subject: Re: [mkgmap-dev] splitter r325: improved split algo and new option


  
    
  
  
    Well, yes - I mainly tried on Austria now, so results may differ a
    bit by country - but for me the trend is clear.

    

    Starting with a .pbf file.

    Fastest is osmcovert it to o5m (about 15seconds) then splitter
    output to o5m. Total time to split 1:11

    Second fastest os osmconvert it to o5m - then splitter output
    osm.pbf =1:14

    Slowest is splitter osm.pbf to osm.pbf = 2:04min..

    

    For mkgmap austria:

    osm.pbf (and creating mapset files twice): 3:32

    o5m (and creating mapset files twice): 3:30...

    

    So for right now, I will output splitter to osm.pbf, but convert
    first to o5m with osmconvert. After successful split, I directly
    delete the o5m file.

    This is using a conventional hdd, maybe with an SSD the o5m
    advantage is a bit bigger (but then you might be more space bound on
    the SSD too)

    The output to osm.pbf by splitter is simply more effective - because
    o5m is 50% bigger. And for Austria the overall time it is faster is
    about 5 seconds...

    

    Of course osmconvert also needs some time (about 15seconds of the
    1:11) - so If I started with an o5m file instead of osm.pbf - it
    would save that 15seconds...

    I only update my maps once a week (and europe continent only every
    ~6 weeks) - so even if I download the planet - I will have to see if
    I maybe cut it to osm.pbf instead of o5m files to save space... I
    don't have time right now to set up all that toolchain however. My
    server is on a 1000Mbit connection - so downloads are faster than
    the HDD if the source is also fast...

    

    For sure converting to o5m before feeding tiles to splitter - makes
    a lot of sense. Even if all else stays osm.pbf..

    

    On 07.05.2014 15:28, Gerd Petermann
      wrote:

    
    
      
      Hi Felix,

        

        yes, confirmed. The output files are much bigger because
        splitter never writes the

        version info (timestamp,uid,user,changeset, version).

        In mkgmap we read the file only once, not multiple times as in
        splitter,

        so the advantage is rather small.

        

        Gerd

        

        > Date: Wed, 7 May 2014 15:20:51 +0200

          > From: extremecarver at gmail.com

          > To: mkgmap-dev at lists.mkgmap.org.uk

          > Subject: Re: [mkgmap-dev] splitter r325: improved split
          algo and new option

          > 

          > Well, I just ran splitter on the .osm.pbf file for Asia
          and austria.

          > True the split is a bit faster to o5m - but the
          difference is not so big.

          > 

          > Compiling from osm.pbf or o5m however - was more or less
          the same speed...

          > on the other hand the o5m splitted files use much more
          space than 

          > osm.pbf (for Austria 293MB vs 450MB - that's about 50%
          more).

          > 

          > I will do some further test with input o5m and output
          osm.pbf vs output 

          > o5m. However if there is also no significant time
          difference (right now 

          > it's like maybe 10% faster output to o5m instead of
          osm.pbf) - then I 

          > think I'm gonna stay splitter output at osm.pbf - simply
          because on my 

          > server I'm a bit harddisk bound (got a large network
          drive for saving 

          > stuff, but with about 20MB/s vs 150MB/s for the local
          drive, it's really 

          > only useful to save stuff there - not for working with
          anything on the 

          > network drive). Splitter input according to the above
          time measures 

          > however - really seems to profit from o5m...

          > On 07.05.2014 12:56, Felix Hartmann wrote:

          > > okay, well the numbers are convincing. I will change
          to o5m and 

          > > osmupdate too - as soon as I find the time (I do
          think I need 5-10 

          > > hours changing my scripts and making sure everything
          is neatly cut, 

          > > but right now I'm too timelimited to do so)..

          > > On 07.05.2014 12:15, Bernd Weigelt wrote:

          > >> Am Mittwoch, 7. Mai 2014, 11:37:58 schrieb Felix
          Hartmann:

          > >>> Well I still use pbf and not o5m.

          > >>> First pbf is smaller..

          > >>> Second - Geofabrik only offers pbf - that's
          why I stayed with it.

          > >>>

          > >>> I don't think I can cut a lot of time by
          first converting to 05m, then

          > >>> hand it over to splitter...

          > >>> Actually I also let splitter output pbf...
          Maybe I could change that in

          > >>> future to 05m..

          > >> I download a planet.pbf, convert it to o5m, time
          ~50 mins,and cut the 

          > >> needed

          > >> polies from that planet.o5m. an update of my
          germany.o5m, 3,2 GB. 

          > >> takes now 3

          > >> minutes. I update the planet once a month, my
          extracts everytime i 

          > >> need them.

          > >>

          > >> thats much faster then download a new pbf
          everyday or update this pbf.

          > >> and with the local planet it is easier to create
          special maps for 

          > >> friends like

          > >> bonn+100 or dach+

          > >>

          > >> Bernd

          > >>

          > >>

          > >

          > 

          > -- 

          > keep on biking and discovering new trails

          > 

          > Felix

          > openmtbmap.org & www.velomap.org

          > 

          > _______________________________________________

          > mkgmap-dev mailing list

          > mkgmap-dev at lists.mkgmap.org.uk

          > http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

        
      
      

      
      

      _______________________________________________
mkgmap-dev mailing list
mkgmap-dev at lists.mkgmap.org.uk
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
    
    

    -- 
keep on biking and discovering new trails

Felix
openmtbmap.org & www.velomap.org
  


_______________________________________________
mkgmap-dev mailing list
mkgmap-dev at lists.mkgmap.org.uk
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev 		 	   		  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.mkgmap.org.uk/pipermail/mkgmap-dev/attachments/20140507/c2129445/attachment-0001.html>


More information about the mkgmap-dev mailing list