[postgis-users] Re: Re: rasters in PostGIS...

Patrick pvanlaake at users.sourceforge.net
Fri Sep 30 00:46:17 PDT 2005


Hi Martin,

"Chapman, Martin" <MChapman at sanz.com> wrote in message 
news:ED3A48B9840E594890A2BC172D11946502F9C0B1 at mailman.san.com...
> Please understand that comments were only meant to offer help.

I'd like to first apologize for my shrill tone in that previous post. It got 
late, I was cranky. Sorry about that.

My main point in the in-or-out discussion is that whether the DB is GB 
without or TB with raster data does not seem to be particularly relevant 
because in both cases there are TBs of data that need loading, analysis, and 
managing. In the enterprise environment where PostGIS is or could be 
prominent (and that specifically excludes data powerhouses like EDC, GSFC, 
USDA, and the commercial outfits) data will be accessed by multiple users 
over a network. So from the point of view of hardware capacity, CPU load, or 
network congestion there isn't too much of a difference.

What really speaks for putting raster data in a DB is the management of 
access and concurrency, and the uniformity of interface. I am not thinking 
of some static data warehousing here, but an active repository that is 
frequently accessed for data analysis. In that case, where multiple users 
are accessing and editing data, a layout that optimizes data access (by 
tiling large datasets), prevents unauthorized access, and guards against 
corruption due to multiple concurrent edits would be more than just a 
trivial feature set.

Mind you that most raster formats are actually not optimally organized for 
the typical analysis task. Whereas many spatial operators use a kernel of 
data around the current pixel (e.g. the slope function) most formats store 
data in rows. This is trivial on small files, but not so on larger files. If 
rasters are to be implemented in PostGIS it will definitely have some form 
of tiling, to optimize data management at the DB level. This opens up myriad 
opportunities for optimized analysis (where an application does not need to 
load large amount of data to work on a small section), mosaicking (e.g. all 
SRTM tiles as a single raster), etc.

Backups and hardware failures? Yes, well, the latter is as much part of life 
as hunger and death and it affects TB databases as badly as GB databases. 
And I have been doing my backups incrementally for the last 10 years and 
although I have never had to backup TBs of data I very much believe that 
PG-Joe will find himself in the same situation.

Even if it is not for the USGS, I think that many users would welcome 
rasters in PostGIS. Read the posts. But doing that right requires critical 
review, so welcome aboard the PostGIS Raster Effort! {;-D

Cheers,
Patrick 






More information about the postgis-users mailing list