> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED]]On Behalf Of marc
> fleury
> Sent: Monday, September 24, 2001 11:13 AM
> To: Jboss-Development@Lists. Sourceforge. Net
> Subject: FW: [JBoss-dev] RE: [JBoss-user] Open letter: ZOAP is dead,
> long live JBossSOAP!
>
>
> an excellent email from Sacha, back to the list (hope you don't
> mind sacha)
>
> point below
>
> |Hello Bill,
> |
> |I just take this off the ML... We will do exactly the same as now. The
> |difference is that instead of keeping container RMI remote
> |references in our
> |smart proxies, we will keep RMI remote reference of the generic
> |RMI MBEAN of
> |our node/Jboss instance + the name of our container. Thus, all
> invocations
>
> correct
>
marc,
I'm just worried that we're duplicating work here. Sacha and I have
implemented RMI replication/fail-over. The stuff we have done can replicate
any RMI object. Thus HAJNDI, Containers, or whatever can use the same RMI
replication features. BTW, this RMI replication stuff is working in CVS,
although Sacha and I think we may have a few JavaGroups issues to resolve.
HAJNDI is working and we'll integrate this stuff with the Containers very
shortly.
> |will go through this RMI invoker and, thanks to the additional container
> |name (i.e. it is a routing information for a JMX node), the invoker will
> |transmit this to the JMX bus.
>
> correct
>
> |As you already told me, you think we will have performance problems with
> |this scheme: you had this problem with Orbix 2000, when a single server
> |socket was overused, it didn't scaled. So we may need to create
> additinals
> |RMI invokers MBEAN "on-the-fly" for some of our containers but the scheme
> |will stay the same.
>
> If we *really* need more sockets then we can open ports on the
> RMI front and
> have new clients talk to these. Say that we limit the number of
> clients on
> a port we can always open new ports. The point is that the client will
> still see the same invoker even though behind the scenes it is a
> round robin
> on independent JMX nodes.
>
Duh! I was thinking we could move the HAPartition idea into the
non-clustered realm and think of a Partition as a POA, but I like this much
better.
> |Are we all in-sync?
>
> I am, you are quite right to stress that it is not a real departure from
> your current design just an abstraction of it. In fact we are
Marc, I think Sacha and I already have the abstraction you're talking about
with HAPartition and our replicated RMI objects. Am I missing something
else?
> talking about
> something I understood during your presentation Sacha... I am
> really looking
> forward to Bill's presentation in Las Vegas as it will be even
I hear some music..."Under pressure. Dom dom dom didda dee doo." :)
> more precise
> and maybe even some code by then (if I get down to it).
>
Bill
_______________________________________________
Jboss-development mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/jboss-development
