Hugues Jerome wrote:
> Marcel Ruff ([EMAIL PROTECTED]):
> 
> 
>>Hi
>>
>>after some discussion with Heinrich, here is a
>>new version.
>>
>>   http://www.xmlblaster.org/xmlBlaster/doc/requirements/cluster.html
>>
>>comments (in any langauge) are welcome
>>
> 
> konichiwa minnasan, daijobu desu ka.

Michele, please translate.

> bon, ok j arrete la

Merci.

> 
> -----------------------------------------------------------
> My two cents regarding this feature (note that I do not know
> the internals of xmlBlaster, so I may be sometimes wrong)
> 
> *) You speak about discovery and lookup, but not about the internal
> configuration. I do understand the lazy login concept to build a tree, 
> but some other concepts are missing : how do I setup my node as a master,
> do you want this caracteristic to be static or not ?

The internal message

   <key oid='__sys__cluster.node.master:heron'>

has a list of message domains which define 'heron'
as a master.

Such a __sys__ message with other domains may be re-published by
a client or an administrator or an domainManagerBusinessObject
which reconfigures the master setup dynamically.

see
http://www.xmlblaster.org/xmlBlaster/doc/requirements/cluster.html
(the example section at the very bottom).

> how do I set 
> certain aspects of the topological tree (like 1. scalability ?)

You are right this is missing.
I will address it.

> 
> (by static I mean that cannot be changed during runtime)
> 
> *) Regarding the features :
> 
> if you look at 'connections between xmlblaster instances', my
> understanding is that if a node do not have the requested message, 
> then it subscribes directly to the master node. this is too restrictive 
> and prevent building a tree structure .. depending how you read 'the'
> 
> take diagramm 1., bilbo will directly subscribes to heron(labelled as
> master), bypassing frodo (labelled as slave)
> 
> hence 'the master' is not precise enough. 

Yes, we need some configuration possibility, thanks for
pointing on it!

> 
> (it seems you implicitly do that in 'routing published messages', but
> stating the obvious is sometimes not bad ;))

Yes again.

> 
> this loop back to the first point, and requires the precise definition
> of 'master' and 'slave' in your context, the way such a relation is
> defined and handled during runtime
> (or at least comment the examples, they seem to provide some useful 
> information I cannot read: __sys__cluster.node.master)
> 
> case 1. and 3. are similar to some multicasting problems, this may
> be a key to a solution

Could you explain further please?

thanks for your valuable feedback

mer lernt jo immer dazue (- Michele please translate.)

regards,
Marcel

-- 
Marcel Ruff
mailto:[EMAIL PROTECTED]
http://www.lake.de/home/lake/swand/
http://www.xmlBlaster.org

Reply via email to