NServiceBus: Pros and Cons of using NServiceBus Distributor

Question

I am considering using a Network Load Balancer to load balance messages between my subscriber instances, instead of using the NServiceBus distributor (which is basically just a software load-balancer from what I can tell). Each subscriber instance will have a queue of the same name for messages to be delivered to, and there will be a virtual IP that round-robins between the subscribers. The publisher will only know about the virtual IP and queue name.

Here is what I understand as the pros and cons of doing this:

1. PROS
   • No need to install NServiceBus Distributor
   • One less thing that would need to be managed/updated when we are scaling-out (we already use an F5 to load balance these machines, and our data center buys know it like the back of their hand)
   • One less point of failure (yes, the NLB could fail, but let's face it, an F5 is going to be a lot more stable than NServiceBus Distributor running on Windows)
   • No need to have a clustered server to have our clustered MSMQ. 2 servers is a lot more expensive than just adding another VIP to an F5.
2. CONS
   • The NServiceBus Distributor allows you to see the backlog of messages more easily since there is a single queue on the Distributor you can monitor. This makes it easy to know when you should add more worker nodes.
   • The NServiceBus Distributor is smarter about controlling of number of worker threads, etc. Gives you more control than an NLB? (not sure about this one)

Have I captured this accurately? I know it is recommended to use the NServiceBus Distributor, and I would like to know more of why before I go against that recommendation.

Pomcroy

Does F5 manage transaction queue load balancing, how are acknowledgement sent back to the sender ? – Sacristan

Answer 1

Youve' got some of the main points down, but one of the main differences is that since the distributor holds on to load itself, if a machine were to go down, the rest of the load would be distributed between the remaining machines with a much lower SLA impact on the messages.

Hartal

You mean because the messages would not be stuck waiting on the node that went down? – Pomcroy