What are the performance considerations when combining a raster analysis server and image hosting server on one machine?

Answer

Along with hosting tiled and dynamic image services based on scalable mosaic datasets, ArcGIS Image Server supports two specialized capabilities—image hosting and raster analytics.

If an organization has resources and limited conflicts in terms of competing raster analysis processing and image service distribution resources, the raster analysis server and image hosting server functionality can be combined and hosted on one server. In this implementation, the raster analysis and image processing operations are executed on an ArcGIS Image Server site configured as the raster analysis server, while the results are managed and published through the same ArcGIS Image Server site. The ArcGIS Image Server site does the work of processing analytics requests, storing the results, and returning results to members in the Portal for ArcGIS site. A schematic of the minimum raster analytics system configuration is shown below.

These two server functions represent inherently different types of demands placed on the server. The raster analysis is more computationally intensive, while image hosting is more related to visualizing hosted services. Having both functions on the same server could result in the following performance issues over time:

• Slow response times when performing analyses on large sets of rasters
• Timeouts for raster visualization services that coincide with the running of larger analytical tasks

The ideal deployment of raster analytics is composed of three server sites to perform the primary roles of the image hosting site, raster analysis site, and dynamic image site.

The advantage of configuring raster analysis with three servers having distinct roles is to maximize efficiency and productivity. Separating distributed raster analysis processing and image service hosting roles enables resources to be dedicated to specific tasks without interruption. In this way, multiple users requesting access to various raster products and services do not compete for and impact raster analysis and image processing resources. The following image shows this ideal configuration.