IHMO, having a Web container running on the same JVM process than EJB container should be an option. It can boost performances, but stucks you in an architecture that is not very easy to manage.
I like the idea to have a EJB server on a specific machine, and other machines around the (internal) net talking to this EJB server. What if you change an EJB? You'll have to change all the Web Container (and you can have many Web container talking to a single EJB container). That means possible mistakes, hat could be avoided by isolating this EJB Container from Web containers. Of course, they are different situation where having a Web contaienr and an EJB container sharing the same JVM process could be the best solution. So : make it optional. BTW, IBM allows you to do whatever you want : WEB container on a JVM and EJB container on another, or both containers on the same JVM, it's just a parameter to set. Emmanuel L�charny > -----Message d'origine----- > De: Alex Blewitt [mailto:[EMAIL PROTECTED] > Date: jeudi 7 ao�t 2003 14:25 > �: Alan Cabrera > Cc: 'Christian Trutz'; [EMAIL PROTECTED] > Objet: Re: Which EJB container? Which Web container? > > > >> I think it should be possible to have many different > >> implementations of the EJB container and the user(admin) > >> should then decide which one he is using ... > > Whilst I think it's a good idea not to be tied down to one specific > system for eternity, I think that Sun's J2EE spec fairly badly split > the EJB and Web containers. > > As a result of which, pretty much every J2EE server thinks of the EJB > container and J2EE container as being distinct. For example, IBM used > to advise running Web containers on one node and EJB containers on > another node. > > I'm a strong believer that the distinction /shouldn't/ be split because > of the terminology, but rather the EJB and Web should be in the /same/ > VM process. It will drastically increase the performance of the > inter-Web/EJB communication since you won't have to use RMI as an > intermediary. > > So, whichever the solutions are, they should be integrated in such a > way that when you run a J2EE application, it runs in a single VM which > provides both EJB and Web services. We should therefore use whichever > containers support co-existence in the same VM, and not go for > implementations which require being implemented in different VMs. > > I also think that whilst some choice is good, too much can be > detrimental. What I'd like out of an EJB server is one that 'Just > Works' and I wouldn't be particularly bothered whether it used > Tomcat/Jetty or OpenEJB/OJB. Nor would I (as an admin) bother about > changing it. > > IMHO it would be better to create a system that works well and quickly > with one set of components (and here, other Apache projects certainly > lend themselves) and not place as much importance whether parts of it > can be changed. > > Alex > >
