[mapserver-users] Ed's Rules for the Best Raster Performance

[Brent Fraser]

I love it!  I had some half-baked convoluted list-of-tileindexes idea; this is much better.  

	We may have to allow PROJECTION=AUTO for the tiles (in the case where the tiles are in UTM several zones), and allow the tile index to be in a different SRS (e.g geographic) than the tiles (I can't recall if this already implemented; I know it caused me a problem some years ago).

A elegant enhancement with great potential...

Brent



Steve Lime wrote:
> Interesting idea. This could take the form of a tile index with a couple of additonal columns, minscale and maxscale. Tiles would
> be grouped together by using common with those values. You could do interesting things like have high resolution data in 
> some areas with other areas covered with lower resolution data over a broader range of scales. The whole layer could have it's 
> own floor/ceiling but tiles would be handled individually.
> 
> I wouldn't handle this as a new layer type, but rather by introducing parameters to indicate which columns to use, kinda like
> TILEITEM. Your pyramids would be defined in the tile index...
> 
> I think the gdal tools would already support this nicely since they add to an index if it already exists so you could run those tools
> over mutliple datasets. Vector layers could also be handled this way, no reason you couldn't have 1 tile per scale range.
> 
> Steve
> 
>>>> On 9/15/2008 at 2:18 PM, in message <48CEB4F5.9020809 at geoanalytic.com>, Brent
> Fraser <bfraser at geoanalytic.com> wrote:
> 
>> Jeff McKenna wrote:
>>> my quick comments:
>>>
>>> - adding overviews with GDAL's 'gdaladdo' utility is very important
>> In some cases.  It depends on your data.  As Ed once posted, it may be a 
>> good idea to switch to external overviews and merge some of the files to 
>> limit the number of file-opens Mapserver must do when the view is zoomed way 
>> out.
>>
>>> - I find your use of the word "pyramid" confusing (this seems to be a 
>>> word that Arc* users are familiar with but has no real meaning in the 
>>> MapServer world...I guess I've been on the "good" side for too long ha)
>>>
>> Not being an Arc* user I can't comment on it's origins.  "Overview" is a 
>> good alternative, but it doesn't seem to convey the same "multi-levelness" as 
>> pyramids.  For example to be able to display the Global Landsat Mosaic, I 
>> created seven levels of a pyramid (each an external overview of the 
>> higher-resolution below it).  
>>
>> Hmmm, may be the plural of overview is pyramid... :)
>>
>> Hey Steve L., maybe we should have a "PYRAMID" Layer type, to replace a set 
>> of scale-sensitive TILEINDEX Layers (this would help my 
>> every-layer-is-exposed-in-WMS problem too: 
>> http://trac.osgeo.org/mapserver/ticket/300).
>>
>> Brent
>> _______________________________________________
>> mapserver-users mailing list
>> mapserver-users at lists.osgeo.org 
>> http://lists.osgeo.org/mailman/listinfo/mapserver-users
> 
>