There have been many discussions on a variety blogs and forums about the necessity of a CAS in a ConfigMgr 2012 infrastructure. The general rule is that if you are approaching 100,000 managed devices or are over that number, then a CAS makes sense. If your estate is smaller than that (and most are), then a single primary site is sufficient to manage your computers.
Now that we are looking at using a single primary site, are we staring a single point of failure right down the barrel? The answer is “Maybe”.
For a small environment, it may make sense to install all ConfigMgr roles and databases on to a single (sufficiently sized) server, but this introduces some risk as configurations like this normally include the package source on the server as well. If this server were to go down, you’d need not only your ConfigMgr backup, but the system backup to get the package source restored.
One way to help distribute the workload in a ConfigMgr infrastructure is to move the customer facing roles off of the site server. Build the site server with enough horsepower to handle SQL and SQL Server Reporting Services. Keep the Site Database, Reporting Services Point, Endpoint Protection Point roles on the site server. Plan for enough storage on the site server to store the ConfigMgr backup locally, but then use your enterprise backup software to backup the ConfigMgr Backup folder.
Move roles like the Management Point and Distribution Points on to other servers. Because the sender is now used to distribute content to distribution points, it is OK to locate a distribution point remotely.