Large networks of brokers, dynamically included destinations and temporary queues

[Eric-AWL <er...@atosorigin.com>]

Hi every body

I want to create a large configuration with dozens of servers. I would want
to distinguish groups of servers by role and assign them static queue names
prefix. To avoid large number of hops, I imagine creating several different
network of brokers connections which will be associated by multicast groups
(with a good dynamicallyIncludedDestinations configuration)

http://www.nabble.com/file/p19070151/Nabble.jpeg 

I explain : when I am connected on a broker of group "A*" and want to
produce a message for Queue "B*", I send it to my broker which knows that
all messages for Queues B* must be dispatch to a group of brokers which
manage B* consumers. And so on for groups C, D, E, F .... (a sort of full
web topology)

each group has a dedicated network of brokers for each prefix of static
queues.

Some of my messages must have an answer, and I imagine using temporary
queues, consumer on them and JMSReply for answer. 

With this configuration, is there a risk that the answer which must be sent
to the temporary queue, doesn't use a direct connection and follow a long
way to find the requestor ?

In such configuration, is it better to use duplex network connectors or two
different connectors ?

I can use ActiveMQ 5.1 if it is better.

Thank in advance
Eric-AWL
-- 
View this message in context: http://www.nabble.com/Large-networks-of-brokers%2C-dynamically-included-destinations-and-temporary-queues-tp19070151p19070151.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.

Re: Large networks of brokers, dynamically included destinations and temporary queues

[Eric-AWL <er...@atosorigin.com>]

Hi Joe.

We were in contact few months ago .... My big project is begun and we have
to find the best topology and test it

when I look web site "Performance" : 

Performance differs greatly depending on many different factors

the network topology 
transport protocols used 
quality of service 
hardware, network, JVM and operating system 
number of producers, number of consumers 
distribution of messages across destinations along with message size 

For us, it is possible to segment our architecture by role and I would want
to use this possibility to imagine the best topology. If I have more than
1000 servers on two sites, I don't want that answer jumps over these 1000
servers and cross many times over MPLS networks, before reaching requestor.

I can imagine creating a pool of answer queues with same prefixes and using
them , but it's not as simple as it appears because I have to be sure that
one answer queue name is not used by other servers which have the same role.
Using Temporary queues with only name over complete architecture seem to be
a better solution.

Eric-AWL

Hi Eric,

As long as there exists the appropriate forwarding bridge(s), w/correct
networkTTL, from the consumer's broker back to the producer's broker, there
shouldn't be a problem with using the request/reply model and temp queues. 

I recommend moving up to 5.1. 

There have been some issues surrounding 'duplex' connections; you may want
to check the JIRA for their status. 

Joe
http://www.ttmsolutions.com  



-- 
View this message in context: http://www.nabble.com/Large-networks-of-brokers%2C-dynamically-included-destinations-and-temporary-queues-tp19070151p19083163.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.

Re: Large networks of brokers, dynamically included destinations and temporary queues

[Joe Fernandez <jo...@ttmsolutions.com>]

Hi Eric,

As long as there exists the appropriate forwarding bridge(s), w/correct
networkTTL, from the consumer's broker back to the producer's broker, there
shouldn't be a problem with using the request/reply model and temp queues. 

I recommend moving up to 5.1. 

There have been some issues surrounding 'duplex' connections; you may want
to check the JIRA for their status. 

Joe
http://www.ttmsolutions.com  



Eric-AWL wrote:
> 
> Hi every body
> 
> I want to create a large configuration with dozens of servers. I would
> want to distinguish groups of servers by role and assign them static queue
> names prefix. To avoid large number of hops, I imagine creating several
> different network of brokers connections which will be associated by
> multicast groups (with a good dynamicallyIncludedDestinations
> configuration)
> 
>  http://www.nabble.com/file/p19070151/Nabble.jpeg 
> 
> I explain : when I am connected on a broker of group "A*" and want to
> produce a message for Queue "B*", I send it to my broker which knows that
> all messages for Queues B* must be dispatch to a group of brokers which
> manage B* consumers. And so on for groups C, D, E, F .... (a sort of full
> web topology)
> 
> each group has a dedicated network of brokers for each prefix of static
> queues.
> 
> Some of my messages must have an answer, and I imagine using temporary
> queues, consumer on them and JMSReply for answer. 
> 
> With this configuration, is there a risk that the answer which must be
> sent to the temporary queue, doesn't use a direct connection and follow a
> long way to find the requestor ?
> 
> In such configuration, is it better to use duplex network connectors or
> two different connectors ?
> 
> I can use ActiveMQ 5.1 if it is better.
> 
> Thank in advance
> Eric-AWL
> 

-- 
View this message in context: http://www.nabble.com/Large-networks-of-brokers%2C-dynamically-included-destinations-and-temporary-queues-tp19070151p19072871.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.