I am trying to optimize raster display with ArcGIS 10.3.1 Server images services.

I have a number of images (250) in tif format which I have displayed using a File geodatabase raster mosaic. According to the size of the Geodatabase (35Mb) it seems that the mosaics are referenced (Do mosaics always reference data or can they have embedded data too?)

There are now various optimization tools but I have really no idea which are relevant regarding rendering speed for services.

What are the differences between Pyramids, Overviews and caches when dealing with image data in a mosaic and which (if not all) should I use to optimize arcgis server images services?

  • 1
    You are correct in your assessment of the Raster Mosaic, it's only referencing the Tif images.. an image catalog can have either managed or unmanaged but suffice to say that a Mosaic Dataset is superior and not much advantage can be achieved by converting the rasters to fGDB format.. Overviews are well worth building as they create mosaic pyramids inside the database that can be shown at bigger scales but I don't know if the individual pyramids of the tif files are used at all if the raster is served as a mosaic dataset... I've heard that caches are worthwhile but have no metrics to support it Dec 17, 2015 at 23:58

2 Answers 2


These names are essentially the same concept, but at different levels of implementation:

  • A pyramid is a feature of a single image, using ~33% more storage (a finite series of 25% smaller images) to capture multiple levels of resampling in a preprocessing step, so an image at a certain pixel size can be generated quickly.
  • An overview is a resampled mosaic of images, preprocessing a collection of image tiles for export at targeted scales. The Image Server Extension of ArcGIS Server can serve overviews directly as an image service
  • A map service cache is an image archive of pre-rendered maps, which may be just a collection of images, or a catalog or mosaic of images, or a combination of those mapping elements with additional symbolized vector data.

Performance is usually dependent on the level of effort a server must expend to respond to a request, where effort is measured in both I/O and CPU. A map service dependent on live queries from a simple catalog without pyramids would require the most effort (heavy load in both I/O and CPU), while a basemap image service can be very fast, or merely fast if it is configured to perform analysis on the fly (e.g., hillshade on a DEM overview). Map service caches generally require the least effort to serve, but only because every possible tile has been pre-rendered and stored in a compressed archive for optimal retrieval.

There are so many options because there are many ways of managing imagery, so if the execution time for using one caching technique exceeds the time between image snapshots, you can use a different technique which results in less overall work.

  • 1
    Or...all three methods together is also possible. Dec 18, 2015 at 6:07
  • 1
    Not at the same time, since the pyramid is used to build the overview, but once that's in place, then the base image isn't used again. The same with a map service based on a catalog or mosaic dataset.
    – Vince
    Dec 18, 2015 at 6:16
  • 1
    Does then an "item cache" do the same job as the "service cache"? Dec 18, 2015 at 11:18
  • 1
    What's an "item cache"?
    – Vince
    Dec 18, 2015 at 12:10
  • 1
    In ArcGIS 10.3.1 > right click on Raster Mosaic in FGDB and you´ll see "build item cache" Dec 18, 2015 at 12:52

I found this blog post comparing overviews and mosaics to be very helpful! https://www.esri.com/arcgis-blog/products/arcgis-desktop/imagery/overviews-and-pyramids-part-1-of-2-what-are-they-and-why-do-i-need-them/

  • 1
    Your answer could be improved with additional supporting information. Please edit to add further details, such as citations or documentation, so that others can confirm that your answer is correct. You can find more information on how to write good answers in the help center.
    – Community Bot
    Jan 25, 2023 at 20:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.