We currently host an API on 3 servers and use NodeBalancer to loadbalance them. One issue we're having is that we can only access it via the public interface, even though we have multiple services in the same datacenter using our API. Performance is crucial and we depend heavily on the response time of the API to keep our service performing to par with our clients expectations. It would be great if we could access the nodebalancer via a private interface, therefore our traffic stays within the datacenter and we can reduce the latency of each request going through our system.

Hello,

While I'm not entirely sure what exactly you're trying to balance, or achieve, I'd like to take a moment just to clarify that even though you're hitting the NodeBalancer on the public interface, the connection goes through our router and so it would never leave the datacenter.

As you can see from this MTR report, a temporarily created NodeBalancer on one of my test Linodes never leaves the datacenter when hit via the public interface:

HOST: localhost Loss% Snt Last Avg Best Wrst StDev
1.|-- router1-nac.linode.com 0.0% 10 0.5 0.5 0.4 0.6 0.1
2.|-- nb-23-92-23-119.newark.nodebalancer.linode.com 0.0% 10 0.4 0.4 0.4 0.4 0.0

On the other hand, when traffic is coming into your NodeBalancer from the external network, the NodeBalancer handles all traffic via the private network to connect to your Linode, so again latency is minimal, if not non-existent.

Please let me know if there's anything we can clarify.

Thanks!

Dave.

He wants the NodeBalancer to have a local v4 address on the front end, so they can balance local things for local things.

If you use the NodeBalancer's IPv6, our datacenters each have their own v6 subnet which our infra and Linodes are numbered out of. In other words, if you used the v6 address locally, it won't even hop through the router since it'll all be layer 2. So, exactly same benefits of using a local v4 address.

-Chris