Best Practice for Optimizing Mosaic Dataset

[TraceStanford1]

I have 2 .SID files that I need to combine into one Mosaic Dataset. One is a county-level NAIP dataset, about 2 GB in size. The other is a 3-inch resolution dataset that we have flown and processed by a private company. That file is about 45 GB. I've combined them into one Mosaic Dataset, but the rendering speed is very slow. Is there a chain of the modification and optimization tools I should run in order to make the dataset faster? I've changed to compression to JPEG with a level of 80, but it is still slow to pan and zoom. I've tried running the Build Pyramids and Statistics tool, but it ran for over 12 hours and did not complete. So I'm wondering if anyone has any advice on how best to optimize performance for this Mosaic Dataset. I'd like to avoid caching it, since it would take up so much space on my server

[PeterBecker · Accepted solution]

MrSID files include pyramids and there should be no need to create overviews (unless you have a large number of MrSID files). If one of the files fully overlaps the other and has a different resolution then you may want to review the MinPS, MaxPS values so that only the appropriate image is being accessed depending on scale. MrSID files should display fast if they are stored on local storage. They are not cloud optimized in that reading any section of the image may required multiple requests to the storage system. Please check reading the MrSID files directly in ArcGIS Pro from the same machine as the server. Are they both slow? The performance for serving should be very similar. 

[JasonMcKeefry]

Spent many hours of troubleshooting poor performance when using mosaic datasets in ArcGIS Pro.  For me, the single biggest factor in this was where the file geodatabase is stored that contains the mosaic dataset.  There is a huge performance boost when the file geodatabase containing the mosaic dataset is on a local drive.  The data that the mosaic dataset points to can be on a network drive.  If the file geodatabase hosting the mosaic dataset is on a network drive, the performance is so bad as to be unusable taking like 20 seconds just to display the smallest file from the most zoomed out overview file.  Once the host file geodatabase is on a local drive, it takes only a couple seconds to display the same.  In our case, the imagery and the mosaic datasets really never change.  The small host file geodatabase is pretty small and can be easily placed on users local drives, while the raw imagery and overviews can reside on a network drive.  The paths stored in the mosaic dataset are UNC paths pointing to the shared network folder that hosts the raw imagery and overviews in our case.  Hope this can save others some time.