Would you, at present date, use JBoss or Glassfish (or another) as Java EE server for a new project?
If you started a new Java EE project today which is to be finished in about a year, which application server would you choose and why?
Part of your answer should include your arguments for your decision. And also how much experience you have with the Java EE server you choose and with the other available servers on the market. These are interesting as we all get a sense of the investigation and thought that was put into your answer.
Answer
I have used WebLogic, WebSphere, JBoss, GlassFish, Resin, Jetty, Tomcat, and a few others over the last 10+ years. So, if I were considering a new project, I would ask myself a few questions first. One thing that I would not question anymore is that I would flat refuse to use JSPs unless I was tortured until I cried for my mommy.
Do I have to be compatible/deploy to a specific product because of someone's mandate? Is there no way to ignore them or convince them otherwise? If so, there's your answer.
Do I have to use EJBs? Really? Avoid them if at all possible--they are really only needed for very large, enterprise-class systems. Remember that they are merely tools, and big ones at that (can anyone say "Golden Sledgehammer"?). They are heavily overused, so really, really question whether you need them. If you do need them, then that removes several of your options including my favorite, Jetty.
Do you have to use any of the other major J2EE technologies like JMS, ESB, etc? If so, and you really can't do without, then you are again constrained to a full-blown J2EE container. Carefully think and investigate before you commit to BPM, for example, and avoid AquaLogic BPM at (almost) all costs--it is ugly in the extreme.
If you really must use a full-blown J2EE container, consider open-source first because it is more robust, better supported, and more cost-effective. They have larger customer bases and more open support interaction, so they tend to get better fixes faster. However, Resin is immature and I would avoid it relative to GlassFish or JBoss--I found it problematic to deploy and support. I would prefer JBoss because of its wider customer base, maturity, etc. GlassFish is harder to incorporate into an automated build/deployment process, but it might be nicer for some of its specific features (if you need them).
Do I have a special reason to need Apache? Then lean towards Tomcat, perhaps plus something.
Can I make do with just servlets? Then I would use Jetty--it is the lightest, fastest, easiest, most flexible solution. If I am leaning against being able to use Jetty, I would question all my assumptions of why. YAGNI applies.
Best is to use StringTemplate/WebStringTemplate on Jetty: a clean, robust, fast, maintainable solution with no licensing fees, solid reputation and support, etc. That is where I start nowadays.
Most applications/systems choose lots of fancy J2EE features when all they really need is servlets and JDBC with some decent architecture/design. Question why you think you need more.
Of the full-blown containers, I would avoid WebLogic and WebSphere unless you are supporting a MAJOR public website (my current employer's website is deployed on WebLogic and it gets eleven+ million hits per month, others have been comparable). WebLogic's real claim-to-fame is their relatively easy clustering, but avoid their proprietary vendor-lock-in features at (almost) all cost. WebSphere is simply a nightmare that I would avoid literally at all cost--I refuse to do projects involving WebSphere after having done a couple in the past. Neither product is worth the massive licensing fees, unless you truly have a special need that drives the use of a proprietary feature. In a decade as a senior architect/engineer for lots of Fortune 500 companies, I have yet to see such a need. On the other hand, I have seen LOTS of pain due to picking such proprietary products.
Even for the really large, high traffic, public websites, the proprietary products are still questionable. I would rather spend that multi-million dollars per year of licensing fees on some good hardware and some quality time from a handful of really good consultants to address a simple scalability solution. The extra millions per year could then be used to produce something worthy of selling on that nice website...
EDIT: another piece to consider...
I have recently encountered Terracotta. I am rethinking everything, and looking to deploy it in a significant system soon. In particular, Terracotta does clustering better than anything else, so I would NO LONGER recommend WebLogic for its clustering.