Hi Nabeel,
The answer to your questions is dependent upon several things. First, when using advanced availability, the recommendation, and what most customers do, is use a hardware load balancer such as an "F5" to distribute the user traffic among the app servers. Most customers have at least 2 app servers, sometimes more depending on the needs and load. Along with that, most customers use a DNS alias to a "VIP" or Virtual IP for the load balancer, which then in turn directs the user's session to one of the available app servers - which partially answers your first question to some extent. The answer to your first question is that typically you would send your users to the App server(s). Now this leads us to answer your second question - again, in a typical AA environment, there would be at least 2 app servers, both behind a load balancer. If one of those applicaiton servers goes down, either by way of some failure, OR by an administrator "quiescing" the server (telling it to send a message to users that the system will be going down, and to save their work and log out in X number of minutes), you would need to manually pull that server from the load balancer rotation so that the load balancer doesnt send more sessions to it. There isnt an automatic way to do this at this time. So to answer your question directly, if an app server is down, and you hit the URL for that app server, it will not present the application - the end users would get an error page. Hence the reason for using multiple app servers behind a load balancer, and having the users hit a DNS alias for the Load Balancer "VIP" so that its seamless to the end user as they are not hitting a specific server directly.
I hope this helps to clarify the functionality of AA, and answers your questions. If you have further questions, please let us know
Thanks,
Jon I.