Hi,

I took some numbers from the real system. We use a custom tile schema with 15 
resolution levels
between 0.5 and 8192 meters. Six of the resolutions levels 0.5-16 m) come from 
the same original aerial images so it would be possible to build lower 
resolution level from previously generated tiles. Nothing special here, 
basically it is the same process than I have done for all the originals by 
running "gdalddo 2 4 8 16 32"  so I believe that the image quality would be 
comparable.

There are roughly 24 million tiles on the disk for the 0.5 m resolution and 
with 5 by 5 sized metatiles it has
taken nearly 1 million WMS requests to seed the cache. There is no alternative 
for seeding this resolution level, it just has to be seeded from WMS and we 
want to have a completely seeded cache. The interesting question concerns 
seeding the 1 m resolution level. In our existing situaltion we have sent about 
250000 WMS requests and split the metatiles into 8 million final tiles. It 
would be nice to know how long it would take to build the tiles by doing 8 
million times "take four tiles, combine and resample". Because there are only 
86400 seconds in a day even small differences count. Even if combining tiles 
would be somewhat slower it might be a good alternative because the WMS server 
would take 250000 requests less and it is a production server which has other 
things to do as well.

In our case the source data of WMS suits very well for seeding because the 
native resolution is the same than the final one and images have overviews 
which also fit perfectly, and the native imagery and cache are using the same 
projection. Obviously systems where WMS server must to reprojecting and adjust 
pixel size would get more advantage from aggregating higher resolution tiles.

-Jukka Rahkonen-


________________________________
Lähettäjä: Puneet Kishor [[email protected]]
Lähetetty: 9. kesäkuuta 2013 20:22
Vastaanottaja: thomas bonfort
Cc: Rahkonen Jukka; Mapserver-Users ([email protected])
Aihe: Re: [mapserver-users] Creating lower aeriel image cache from the best 
resolution tiles

Yesterday at SOTMUS here in SF, Dane Springmeyer (Maobox) demonstrated 
something they call "overzooming" that might be worth investigating. It seems 
to solve the kind of problem Thomas is articulating.



--
Puneet Kishor
Policy Coordinator for Science and Data
Creative Commons

On Jun 9, 2013, at 9:14 AM, thomas bonfort 
<[email protected]<mailto:[email protected]>> wrote:

Jukka,
No this is not considered, and you'll have to be really persuasive to make this 
happen as I do not see any advantage of adding this :) The reason is that in 
order to create e.g. the tile for level 0 for the first time, you'll have to 
downsample (and maybe previously seed) the full z=21 level (and all the 
intermediate ones).

regards,
thomas


On 9 June 2013 18:02, Rahkonen Jukka 
<[email protected]<mailto:[email protected]>> wrote:
Hi,

Have you been considering to seed the cache for aerial or satellite image 
layers in opposite order compered to the normal Z0, Z1, Z2... so that only the 
best resolution Z(max) tiles (or perhaps also Z(max-1) were requested from the 
WMS? The lower resolution tiles could me created by combining and resamplint 
tiles from the tiles of the higher zoom level in groups of 4. This should be 
much faster than asking lower level images from the WMS, especially if the WMS 
is a remote one and connection is slow. If resampling from Z+1 tiles yields to 
artifacts I feel that quality by resampling from Z+2 would normally be good and 
speed migh be still fantastic.

-Jukka Rahkonen-

_______________________________________________
mapserver-users mailing list
[email protected]<mailto:[email protected]>
http://lists.osgeo.org/mailman/listinfo/mapserver-users

_______________________________________________
mapserver-users mailing list
[email protected]<mailto:[email protected]>
http://lists.osgeo.org/mailman/listinfo/mapserver-users
_______________________________________________
mapserver-users mailing list
[email protected]
http://lists.osgeo.org/mailman/listinfo/mapserver-users

Reply via email to