On 8/11/06, Eugene Prokopiev <[EMAIL PROTECTED]> wrote:
> If you have a clustered JDBC database and want to use that to get HA
> then I'd recommend using this...
>
> http://incubator.apache.org/activemq/jdbc-master-slave.html
>
> which allows you to run many brokers; one is the master and others are
> slaves for fast HA and failover. The only downside is performance as
> JDBC Master Slave only works when not using the high performance
> journal
I know about this limitation but I need network of brokers instead of
master/slave for parallel messages processing.
So a single broker (or single master/slave) can support many clients
(500-10,000 depending on your OS) and deal with large amounts of
concurrent message processing. You get parallel message processing via
having many JMS clients (producers & consumers); a single broker can
support that fine.
Perfomance is not a
problem in my case because I have another points of slow processing.
So, my question is the same : can I got any advantages by setting up
network between brokers or shared database is enough?
The only real benefit of networks is if you need store-and-forward
such as if you have a slow network link between producers and
consumers. If clients are arranged randomly, networks can load balance
traffic across multiple brokers - but its often rare to have so high
traffic that you need more than one broker - its usually the client
side which is the thing you need to run lots of.
--
James
-------
http://radio.weblogs.com/0112098/
