Many times when on site I'm asked how do we achieve "guaranteed delivery" or "zero message loss"? with Neuron. The answer is somewhat involved but hopefully after reading this blog you'll understand the pieces that need to be in place.
First, is the environment 2003/XP? If it is, then the story is that if all messaging i.e. parties will run on the server then you run one Neuron server and use the MSMQ channel. Cluster the Neuron box for HA.
The only scaling that is available in this configuration by buying a bigger box...
If you have clients i.e. remote parties that communicate with the server in a 2003/XP environment then you must also use Neuron's server cache mode to prevent message loss.
If the environment is 2008/Vista then you gain some flexibility. You can now move the MSMQ server to a separate box. You'll want to cluster it for HA. Then you can point multiple Neuron servers using the MSMQ channel at the MSMQ box for scaling. MSMQ 4.0 supports remote transactional reads so you do not use server cache mode in this configuration.
One of the hot debates within the Neuron team is should we provide another durable and transactional topic channel say based on a database so that we could enable more flexible farm scenarios...
If you think this would be valuable, I'd appreciate comments or go ahead and start a thread on the forum!