Mosaic Dataset/Image Service Performance

[BruceBurwell]

Hello

I have an orhorectified Geoeye mosaic dataset that I have create a service from and the performance is very slow (roughly 10 seconds to render at 1:9000 scale). The mosaic dataset is part of a map service and has a scale reference set up for 1:9000. I have created pyramids for the data. Its being orthocorrected and colour balanced on the fly and reprojected to web mercator from UTM. I am just wondering if there is anyway to speed up the rendering without caching the data. Does anyone have any tips?

Thanks
Bruce Burwell
GIS Analyst
Saudi Aramco

[JeffreySwain]

So you have a mosaic dataset in a map service, but do not want to cache it.  Is that correct?  Generally most people that have them cascaded are in order to create a map cache. How does the image service do on its own?  Cascading the image service in a map service will not be any faster than the image service.  If the image service is slow then test by publishing one overview directly and one of the source rasters directly. If those are fast then it is the way that you have configured the mosaic dataset.  If both the original raster and overview is just as slow then it is the network speed. At this point I would determine if you move the source rasters to the Server and they perform better or work with your network professionals to determine why the network is causing the problem. 

I would recommend creating more overviews on the original mosaic dataset, which will take advantage reducing the threshold at which the source rasters are being utilized, which may speed up the performance as the overviews are quicker to display.   Depending on the original raster format, some rasters are just slower performing.  Generally the more highly compressed then the worse the performance.  There was information generated by testing the original Image Server, but in tests here the same holds true in testing here.  JP2s and Sids perform slower than the fastest performer, tiled tifs.

[GordonSumerling]

Bruce,

I have seen performance increase if the number of vertices in the boundary polygons is kept simple. That's to say when you build boundaries keep the vertice count low.