For any service you publish on ArcGIS for Server, one or more instances will start on the GIS servers to represent that service. Each instance takes resources from the machine it is running on. The number of instances on each server can be configured when you publish the service. Each GIS service differs in terms of memory usage and processing consumption, and the same thing applies on the GIS servers. You might have different generations of servers with different specs and resources, so it makes sense to have some sort of distribution window by which you can specify which services will run on which servers. To manage this in an efficient way, Esri came up with a technique to group GIS servers into clusters and then permit you to configure which service goes to which cluster of machines. Clustering is an advanced technique that can prove to be of use if configured correctly. For instance, you have some unused workstations or some standard-issue PCs lying around in your inventory...