Pyramids - are they used by image server?

[MarkPeacey]

I'm fairly new to image server (v9.3.1) but I'm wondering if normal pyramids (.rrd files) are used by image services? I gather service overviews, derived tiles and rasters have some sort of pyramiding functionality, so I can't see there is a need to keep the .rrd files if the imagery is only served up by image server.

[PeterBecker]

The RRD files are used if you add the data using the 'ArcGIS Raster Datasets' raster type.  Pyramids are only of use for larger images (>7000 cols). Only the pyramid level were the Number of Cols is >1500 are added. IE if  you added data that has 10,000 cols and a RRD file, then the the first level (5000 cols) and second level (2500 cols) will be added but not the 3rd level (1250). This is so that the system does not try and use the overviews from lots of images at smaller scales. RRD files are not quite so optimum as a format and are not compressed so are relatively large (they are the same format a .IMG). If you want to optimize these and using TIF as a source then I would currently suggest using internal pyramids. These can be created using GDAL gdalaaddo command from FWTools. This can be directly used by ArcGIS Desktop as well. Note that in V10 OVR files (which are TIF format and can be compressed) are used as the default method for generating pyramids of raster datasets if none exist. IE having internal pyramids will be used by 9.3 and V10. Pyramids have most value when the images are large.

[DallasShearer]

If you create internal pyramids on the generated "service overviews" and/or "derived tiles", would image server use these pyramids also? Given they adhere to the "Number of cols" rules stated previously?

[PeterBecker]

Yes the technology uses any existing pyramids. The system does not necessarily use all the levels though. When a rasters is added the system computes the LoPS and HiPS values for the rasters. The LoPS is based on the high res imagery, the HiPS is based on the availability of pyramids and their size. Generally the HiPS is set to the size of pixel in the pyramid where the NCols>1500. What this attempts to do is ensure that the larger pyramids are used by not the very small ones. At the larger scales is it better to create overviews that are a mosaic of multiple images so that the system does not need to open a large number of files at smaller scales. All these value can be manually changed by using the Calculate tools.
V10 is better then V9.3 in that the overviews are automatically used if they exists. In V9.3 when ingesting the imagery the system looked to see if the overviews exist and if they were added later it would not recognize them.

[MelanieHarlow]

If you create internal pyramids on the generated "service overviews" and/or "derived tiles", would image server use these pyramids also?

Please do not build pyramids on your service overviews. This is not recommended. These are generated within specifications that do not require pyramids. (You can think of them as pyramids for your ISDef)