I would count JDMK out, I've used it and it's basically JMX + a few classes
for proxying mbeans from one agent to another. I used it originally to code
my mbean watchdog service (which I'm gonna post as a zip to the board soon),
I was able to recode my stuff to use the org.jboss.jmx.* stuff very easily.
I think Sun dragged loads of code out of JDMK to create the JMX-RI, or
vice-versa.


-----Original Message-----
From: Rickard Öberg [mailto:[EMAIL PROTECTED]]
Sent: 14 March 2001 07:30
To: JBoss-Dev
Subject: Re: [jBoss-Dev] Clustering based on X?


Hey

> Jini
> ----
> + Allows highly dynamic cluster architectures
> + Removing and adding a service (EJB service) is easy
> + Group architecture
> + Failover could be implemented very efficient
> + Cluster components can find each other using multicast discovery
> + Provides already a TM

I don't think we'll have use of the TM, but I'm not 100% sure on that.

> - A lot of features & parameters could be involved (Leasing,
EventHandling,
> GroupConfiguration, etc.)

I've done some work with the Jini 1.1 tools, which has nice little classes
to abstract away most of the hardcore stuff. You can do pretty amazing
things in 10 lines of code.

> - I'm not sure about the stability

I'm only worried about the LUS, since it's not manageable (at least not the
Jini kit one. There may be alternative impls that are better).

> - Complex behaviour if one starts to implement distributed synchronization
> (e.g. between entity beans)

This has more to do with the algo chosen to do entity synch rather than Jini
itself.

> - Network Traffic
> - A lot of different daemons must be started, when using Jini to the
fullest

I've found that only the LUS is necessary. The RMID is not really necessary.

> JDMK (commercial)
> -----------------
> + I think this would be the best technology for clustering:
>    The combination of a JMX with more functions and a Jini with less
> functions including multicast discovery
>    (That's how it looks to me. I know why Sun sells it...)
> - It is commercial and costs something

Not "something"; A LOT. Not an option I'm afraid :-(

> Pure RMI
> --------
> + Standard, that is not going to change soon
> - Almost no code and location transparency
> - A lot of work.

Yup. Jini is purely "on top" of RMI though.

> If someone want's to play with a Jini-Cluster and do some further research
> on this possibility, he can visit a Cluster Prototype for jBoss using Jini
> here:
> http://www.fhso.ch/~gasche-m/projects/projekt888/

Cool :-)

regards,
  Rickard




Reply via email to