Hi!
Filip Hanik wrote:
> I too have to agree, soap would be way to slow. And again, the multicast
> messages do not have to involve the web server, only between the servlet
> engines and between the app servers in a cluster.
No, only between app servers in the cluster. We are talking about
clustering the EJB containers here, nothing more. As a bonus it would be
possible to use Jini for accessing JNDI from clients (e.g. servlet
engines) but that's purely a bonus and can be done through direct
specification of Jini Lookup Service host instead of using multicast to
find it.
> are you planning on implementing the same features/restrictions as Weblogic.
> in other words, replicate the entire HTTP session,
This is not something we would do, since we're not doing a servlet
engine.
> create replica aware
> stubs for the home interfaces.
Yes.
> but once your client holds a connection to a ejb object it is considered
> pinned, and it's state will not get replicated unless it is a read only data
> bean.
> because it would be a nice feature, if the fail over was transparent even
> once you hold a reference to an EJB object.
Something like that.
> I'd like to help out, any design docs or anything tangible I can dig in to?
Check BugZilla for clustering.
/Rickard
--
Rickard �berg
Email: [EMAIL PROTECTED]
