I have published a test map service using caching, but I am not seeing any noticeable differences. Are there any performance metrics at which I should be looking/on which I should be focused?
What version of ArcGIS Server are you using to publish the mapservices? I ask because my organization had an application where most of the 10.0 mapservices needed to be cached due to the performance of the mapservices (The web adf architecture might have played a role in the speed of the mapservices rendering as well). ESRI made great improvements to the mapservices by 10.2 so most of these cached mapservices were replaced by a very fast dynamic mapservice which you don't need to worry about updating as the data is dynamic.
My suggestion would be to test the application using dynamic mapservices first to determine if the simpler setup meets your needs.
Thanks for the response, Michael.
We are using 10.4.1., and we are currently using dynamic map services. We are seeing some slow rendering, however, so I was mulling using cached services, since the data itself does not change very often.
Do you think you can lower the scale dependency for the slowest rendering layer(s) any further (e.g. You have your Census Tracts only becoming visible at 1:15,000 now but lower it to 1:8,000 - Just an example as I don't know how you have your scale dependencies currently setup)?
In addition to what Michael is saying, it would be best to prepare the map document like this because if the map draws slow when opening it in ArcMap, chances are that it will render slowly in Server.
Thanks, guys. I appreciate the feedback.
We actually do have the map document set-up so that layers only appear at certain scales. The rendering in the browser is based on user selections on the screen. So, different columns of the map service display depending on user selections and the renderer is based on the rendering algorithm/number of breaks a user selects. Right now, we are investigating all possibilities...network issues, server configuration, etc. Caching was just one option that seemed like a low-hanging fruit.
Have you tried to duplicate the end user selected rendering of the layers in the mxd to see how different rendering scenarios perform in ArcMap? If you have only tested the default symbology in ArcMap, then you have not truly tested what the endusers of your application will have the same layers rendered as. I would suggest testing a good number of different rendering schemes (maybe 5 or 6) to determine the speed of these renderings vs the speed of the default rendering (baseline) you have setup in ArcMap.