A best practice question - how are very large websites best structured with Java.
I'm interested in knowing how the deployments themselves are structured -
Some possible answers:
Is there any documented best practice for this?
If you could find data, I'd bet you'd have examples of each one (and perhaps more besides).
I don't know that there's a "best practice" that's uniformly followed.
Your biggest concerns appear to be session sharing and deployment. Regardless of how it's done, I'd say that session data ought to be minimized, and sharing between WARs? No. One owner for data, please. Sharing suggests that you've spread functionality for a single use case across modules. This will lead to grief someday.
As far as packaging goes, I'd say that the bigger the package the more code is affected by changes. If you can partition something into two independent WARs, you can change one without bringing the other down. That's better for maintenance.
When you say "very large websites", I assume website that have too many modules and sub modules. Not the site with heavy traffic. If you are intrested in site with heavy traffic, please rephrase and post a different question or update this one.
Keep it modular with multiple EAR's/WAR's. Having it modular gives opportunity for each module to evolve independently of other unless there is change in module interface. If there is change in interface, dependent module needs to be updated as well. Although not a website, one of such example is Eclipse IDE. Each of it's modules are developed and maintained independently and has it's own versions. Having website modular also gives opportunity to deploy individual modules on separate machines/servers and hence can scale individually.
Session sharing across EAR's/WAR's will be considered bad idea. It is too much of a work for server. Further it can introduces bugs that will be hard to debug just like global variables. But having user to log in again and again for each module is also really bad. You would need to implement some "Single Sign On (SSO)" solution for this. As far as example is concerned, www.google.com, www.gmail.com, www.orkut.com etc. are all google services and each service is like an individual module. But if you sign into one of the service and then open another without signing out, you automatically get logged in.
Although modules modules that are assembled into one big War is a bad idea, Once a year you can have all modules being deployed together (not as single WAR but individually) and name it something. Similar thing can be seen with eclipse. Eclipse has time to time updates for each individual modules but one a year they have a major release where by all modules are upgraded (Europa, Ganymede, Galileo ...).
Every application is different and has different requirements. There is no specific best practices for large website as it would depend on the website being developed. Say for example session sharing won't be a good practice but business requirement may drive you to do so. Or some alternate method could be used to share information across modules.