so I was thinking...
i have published a map service with ortho imagery for one of the smallest states by area in the US and it takes 10 days to create the cache.
i was wondering if anyone has published service with a much larger state by area? the thought of it taking over a month to create a cache keeps me up at night.
We've done the state of Florida. We had many issues. Time to cache and size of cache definitely contributed, but publishing from a windows environment to linux ArcGIS Server was also a major obstacle. There was a whole process we had to do creating a mosaic dataset and then breaking the file paths and then copying to the linux box because of the file path differences between windows and linux. It's been awhile so I don't remember the exact details, but I believe the time was 9 - 12 days and close to a terabyte of disk space. We had four server boxes assigned to tile creation.
I've cache the state of Alaska at 10 LOD and it took about 8 weeks (v10.2.x). Since then we purchased the Image Server extension, added 4 additional level of detail and (if not for other network issues) it would only take and estimate of about 2 weeks. Being able to use the mosaic datasets greatly improved the speed. But the Image Server extension also eliminates the need for some of the caches since it does faster on-the-fly display.
I would be very interest in this information.
I think the important parameter is the area and the smallest scale.
We created cache down to 1:1000 for about 20000 square kilometer.
I think it took less than a week using 6 processes.
The interest point is when you start the process each of these 6 processes is using a full core (12% in case of 8 core machine) but after a few hours is slowly going down to 2-3% per process.
We use the image extension.
i guess while were on the topic of # of cores....
for a statewide deployment where you have a desktop application calling on services in almost 300 locations with a total of 800 seats statewide...
4 Cores each
2 instances each
I guess i would be interested in knowing if other would find that sufficient of it should be bumped to say something like 8 cores per VM and something like 4-6 instances?
thanks for everyones comments, this have been a great thread so far!