GeoServer: Best way to publish 2500 TIFF or 71 ECW files?

Question

I have ortophotos of an area that I need to serve as a background map for my vectors. I got it as 2500 files (71,5MB each) in raw TIFF format with corresponding TFW word file - 180GB of data. Coordinate system is local and it matches with my vectors (there is no EPSG code for it but I named it as “32805” and put a proper definition).

For desktop use in MapInfo I converted them to ECW (with some tool that comes with MapInfo) and retile as much bigger to have only 71 file because opening 2500 tiff files is overkill. I just merged 49 TIFF’s (7x7) into one ECW - 35000x35000pixels – largest is around 200MB) It works great and its really fast in MapInfo.

Now I am quite confused - HOW to serve them in GeoServer ?

I have publish one TIFF and one ECW for comparison. ECW is much faster in browser preview (I am aware of ECW server licensing issue but that shouldn’t be a problem). I have found a presentation “GeoServer on steroids” and read about ImageMosaic, ImagePyramid, retiling, adding overview etc it was quite informative but still don’t know what to do.

My question is : how should I do it ? Mosaic or Pyramid, and if is one of the answers positive, I need your advice or some hints. I would really like to be ECW because of disk space so there will be no need to keep 180GB of tiff’s on a server.

Data will be served through LAN with maximum 20 users connected at peak hour.SQL server have not so huge amount of data. Sorry if I miss some other info, but I will send it if needed.

---

Geoserver 2.1.4, Windows 7 32bit, 2GB System Memory, (1.7.0_09 (Java HotSpot(TM) Server VM), Native JAI+Native JAI ImageIO = true

---

Original TIFF
gdalinfo D:\75720-47970.tif
Driver: GTiff/GeoTIFF
Files: D:\75720-47970.tif
       D:\75720-47970.tfw
Size is 5000, 5000
Coordinate System is `'
Origin = (7572000.000000000000000,4797500.000000000000000)
Pixel Size = (0.100000000000000,-0.100000000000000)
Metadata:
  TIFFTAG_SOFTWARE=Adobe Photoshop 7.0
  TIFFTAG_DATETIME=2006:10:09 13:02:57
  TIFFTAG_XRESOLUTION=72
  TIFFTAG_YRESOLUTION=72
  TIFFTAG_RESOLUTIONUNIT=2 (pixels/inch)
Image Structure Metadata:
  INTERLEAVE=PIXEL
Corner Coordinates:
Upper Left  ( 7572000.000, 4797500.000)
Lower Left  ( 7572000.000, 4797000.000)
Upper Right ( 7572500.000, 4797500.000)
Lower Right ( 7572500.000, 4797000.000)
Center      ( 7572250.000, 4797250.000)
Band 1 Block=5000x1 Type=Byte, ColorInterp=Red
Band 2 Block=5000x1 Type=Byte, ColorInterp=Green
Band 3 Block=5000x1 Type=Byte, ColorInterp=Blue
-------------

ECW file which is retiled from 7x7  original tiffs

gdalinfo D:\OF-45.ecw
Driver: ECW/ERDAS Compressed Wavelets (SDK 3.x)
Files: D:\OF-45.ecw
Size is 35000, 35000
Coordinate System is:
LOCAL_CS["LOCAL - (unsupported)",
    UNIT["Meter",1]]
Origin = (7571500.000000000000000,4798500.000000000000000)
Pixel Size = (0.100000000000000,-0.100000000000000)
Corner Coordinates:
Upper Left  ( 7571500.000, 4798500.000)
Lower Left  ( 7571500.000, 4795000.000)
Upper Right ( 7575000.000, 4798500.000)
Lower Right ( 7575000.000, 4795000.000)
Center      ( 7573250.000, 4796750.000)
Band 1 Block=35000x1 Type=Byte, ColorInterp=Red

  Overviews: 17500x17500, 8750x8750, 4375x4375, 2187x2187, 1093x1093, 546x546, 273x273, 136x136
Band 2 Block=35000x1 Type=Byte, ColorInterp=Green
  Overviews: 17500x17500, 8750x8750, 4375x4375, 2187x2187, 1093x1093, 546x546, 273x273, 136x136
Band 3 Block=35000x1 Type=Byte, ColorInterp=Blue
  Overviews: 17500x17500, 8750x8750, 4375x4375, 2187x2187, 1093x1093, 546x546, 273x273, 136x136

Answer 1

I ran an experiment with a TIFF file and an ECW. Started with a 1.2 GB ECW, and converted it to TIFF with compression and pyramids, it was ~1.5 GB. So I think that a TIFF can be a similar size to an ECW.

I would mosaic the image using GDAL, ensuring that compression is on. Then build pyramids, and if the resulting file is reasonable (less than 10 GB, I suppose) I'd just let GeoServer do the rest.

Performance between PostGIS and TIFF will favour the tiff, I understand.

References:

• http://linfiniti.com/2009/09/image-mosaicking-with-gdal/
• can't find others...I should document what I do better!

Answer 2

I had a similar problem few weeks ago . I resolved it this way:

1. creating pyramids rasters image (all rasters had got pyramids depends on standard scale rate in my project
2. creating tiles from raster (mosaic)
3. putting all files to postgis (by WKTRaster)

By this way you get MRDB (multi-resolution data base) which is the most effective way of serving a large amount of data.

After above you can just simply connect GeoServer to PostGIS and serve your data. Based on my own example, I had to use 82 ortophotomaps (40GB of data) in my application, so I did as followed and it works great! The disadvantage of this situation is that raster tiles are much bigger than the source ones. So in my case the data grew from 40GB to ~96GB.

EDIT And you should monitor yours server parameters because 2GB of RAM and win7 + geoserver + postgres can choke sometimes. Maybe a good way of increasing performance would be moving the DB to another machine, or changing Win7 to Linux (or both of those) because *nix system are less expensive than a MS one.