Web Hosting Talk : Web Hosting Main Forums : Colocation and Data Centers : Load Balancer Questions - Barracuda 340

[John[H4Y]]

Attached is a .txt file showing my desired topology. Here is my question for any load balancing (and especailly Barracuda) gurus:

I've got 3 machines and I need to balance tcp port 5222 between them using a Barracuda 340 load balancing appliance. The 3 machines are currently connected publicly to a 100Mbbps switch and we want to avoid taking them offline for even a second SOMEHOW. They each have public IPs that we want them to remain accessible by no matter what.

They all have dual NICs and all also have private IPs that are bound via a seperate internal Gbps switch between them.

We want to add the Barracuda device to that private switch, and then give it a SEPERATE uplink to the intenet (a new gigabit uplink). So the internet would be connected to the WAN port on the barracuda, and then the barracuda would be connecting to the internal switch.

The 3 machines may be 192.168.4.1, .2, .3 to the load balancer, but the load balancer itself will have its own public IP (virtual IPs). We would then direct the traffic to that public IP for it to be balanced between the three machines behind the private network, even though we could access the eth0 IPs on the same machines via the route through the public switch at the same time.

Can this be done without problems, if not, how would you recommend changing the topology, and if so, which method of load balancing should I use and any tips to set it up?

[John[H4Y]]

For instance, would the Direct Server Return setup be a good option in this case?

[sct4a]

Originally Posted by John[H4Y]

For instance, would the Direct Server Return setup be a good option in this case?

Yes. Unless you have another reason to use both interfaces and seperate switches.

[John[H4Y]]

So that should work, correct? The other reason is to isolate the load balanced connection vs a non-load balanced direct internet connection but have both available.

[sct4a]

Direct Server Return / Local Triangulation should work no problem. Why would you want to isolate the load balanced connection vs non-load balanced? If you have multiple connections and switches it would make more sense to design it so it operates in a redundancy mode unless there is specific reasons not to.

[John[H4Y]]

We just want to keep it from going offline. If we do it this way, we should be able to point DNS at a new IP (the load balancer WAN IP) and if there is a problem, we can put it right back to the DNS round robin it is currently using between the 3 public IPs. Also, the servers are all using the internal switch/eth1 connections already for communicatoins between web servers and database servers as an example. In other words, I am trying to add the load balancer while keeping the private and public access to these machines unchanged.

ex: server 1 before adding the load balancer has public IP of 199.199.199.1 and private IP of 192.168.4.1 - it is in production and connects to server @ 192.168.4.2 for DB activity. If I set this up with direct returns, I should be able to just plug the Barracuda LAN port into the private switch and a new WAN uplink to the WAN port, correct? Does it matter if it is not the same uplink/switch as the servers are connected to via each of their eth0 NICs?

[sct4a]

You can avoid going offline while also having redundency. It isnt a very good design to have those two split networks just to be able to switch back and forth. I'd use both connections redundently instead of splitting them along with a DSR/Local Triangulation design which you can accomplish what what your looking to do. If you really want two seperate split networks then use regular NAT load balancing method.