We've been hunting for a long time to improve out PI Vision loading times. Load times are ok-ish with all below 10 seconds for even the most crowded displays. Load time is referred to as pressing refresh/F5, and wait for the display to fully load.
One of the issues we still need to resolve, is that sometimes (not always) the initial load time of a display for a user is slow (60 to 90 seconds). The .HAR file shows this is always due to a slow response on the first diff4data call for the display.
Any ideas how to find out what causes this? The issue is not with the actual data, as a subsequent refresh is quick enough, so this is likely related to user-specific caching. Would tracing on AFSDK on PI Vision webserver make any sense?