[GRASS5] v.support and large datasets (gshhs from GMT)

[David D Gray]

Markus Neteler wrote:
> 
> Hi all,
> 
> to provide a high resolution vector coastline dataset I have
> converted the
>                         G S H H S
>  Global Self-consistant Hierarchical High-resolution Shorelines
>  from Paul Wessel and Walter. H. F. Smith
>  (ftp://gmt.soest.hawaii.edu/pub/wessel/gshhs/).
> 
> A problem arises from v.support which doesn't run on large files (at all or
> or in reasonable time, process is still running here). If you are interested
> to test GRASS with a *large* vector file, I have prepared the GMT dataset
> for you:
> 
> To describe it's resolution:
> (From the GMT page:) The file gshhs_h.c (High resolution) is at 0.2 km, I
> have converted gshhs_f.c (Full resolution) at 0.?km (no value given). So I
> have converted the *full* resolution dataset gshhs_f to GRASS ASCII vector
> format for you to download at:
> 
> ftp://www.geog.uni-hannover.de/outgoing/gshhs_f_grassascii.tar.gz (58MB)
>  -rw-r--r--   1 neteler  users    226115131 Dez  8 17:42 dig_ascii.gshhs_f
>  -rw-r--r--   1 neteler  users    10561544 Dez  8 17:42 dig_att.gshhs_f
>  -rw-r--r--   1 neteler  users     2166637 Dez  8 17:42 dig_cats.gshhs_f
> 
> You can use this lat/long world coastline dataset within "global dataset"
> location which you get from here (or you define an empty world lat/long
> location):
> 
> http://www.geog.uni-hannover.de/grass/data.html
> Add the missing directories:
>  $LOCATION/dig_ascii
>  $LOCATION/dig_cats
> 
> and move the files into their directories (rename to gshhs_f or similar of
> course). Then
>  v.in.ascii in=gshhs_f out=gshhs_f
>  v.support gshhs_f op=build
> 
> However, this takes muuucchhh time... Does anyone see a chance to speed
> up the v.support topology build function? Maybe it would be useful to
> add progress output (percentage) to v.support.
> 

I don't know the details but execution time against file size grows at a
rate that is `more than linear' for large files. So large files, like
vector grids, do build very slowly. I think the answer is segmentation.
Perhaps the technique of `walls' where a triangulation topology is built
from a set of points could be modified to build areas. Then only small
regions are built at a time working out from the segment boundaries.
This needs to load the respective segments into a network in memory to
be efficient. A rework of the build function is desperately needed. This
could be in place for 5.1. Anyone else has had thoughts about this?


David

> Regards
> 
>  Markus
> 
> ----------------------------------------
> If you want to unsubscribe from GRASS Development Team mailing list write to:
> minordomo at geog.uni-hannover.de with
> subject 'unsubscribe grass5'

---------------------------------------- 
If you want to unsubscribe from GRASS Development Team mailing list write to:
minordomo at geog.uni-hannover.de with
subject 'unsubscribe grass5'