[Liblas-devel] Fast way to read compressed LAZ files

[Martin Isenburg]

Hello,

by default LASzip compressed points in independent chunks of 50000. Hence
LASzip can be trivially paralized on decompression by spawning multiple
thrreads that each seek to a different multiple of 50000 and then
decompress non-overlapping chunks of multiples of 50000 points.

Of forumated otherwise. Seeking and decompressing a random point in the
file will incur a decompression overhead of 0 to 49999 points depending on
where your target points falls within the chunk of 50000 points.

Regards.

Martin @rapidlasso

On Fri, Feb 5, 2016 at 12:51 AM, Alexander Biddulph <
Alexander.Biddulph at uon.edu.au> wrote:

> I am currently using liblas (with LASzip) to read compressed LAZ files.
>
> My current workflow is to open the LAZ file, read in each point
> sequentially and place each point into a cell in a regular grid, then find
> the centroid of each cell.
>
> However, this process is currently taking an extraordinarily long time, so
> I was wondering of there is some way that I could speed this process up.
>
> I was initially thinking about extracting points from the file using
> multiple threads (via OpenMP). However, based on the small tutorials
> provided, it is my understanding that each thread randomly accessing the
> Nth point in the file would require decompressing the first N points. So I
> was then thinking about decompressing the entire file first, then randomly
> accessing the points in the decompressed file, however, I am unable to tell
> if this would actually provide much of a performance increase over just
> reading sequentially from the compressed file.
>
> Can anyone suggest a good way of speeding this process up?
>
> Thanks,
> Bidski
>
> _______________________________________________
> Liblas-devel mailing list
> Liblas-devel at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/liblas-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/liblas-devel/attachments/20160205/8d51ed89/attachment.html>