Hi @RF_Dude 

Thanks for posting in our business forum.

RF_Dude wrote

Problem Statement:  Two seemingly equal WAN feeds to GW via two StarLink terminals are not sharing load equally.  WAN 2 takes majority of load, while SFP WAN (1) a fraction (<20% when busy and <10% when not busy).  This is consistent over months.  

 

Previous inquiries in this forum exist for this question...  Understood that aggregating or bonding requires specialized balance equipment at both ends.  This inquiry is only about how the "Application Optimized Routing" handles traffic.  Based on the results, seems one WAN port is favored, while the other is used for overflow?  Below are my observations...  

  

Router is set for 1:1 load balancing.  Application Optimized Routing is enabled, as it should be for consistent client application experience.  Link Backup is unchecked (disabled).  Latency shows the same at 35ms for both SL terminals.  

 

Local to GW Starlink feeds WAN port (labelled 2).  Starlink from next building feeds back to GW as a VLAN, untagged, and into the SFP WAN using a copper SFP (labelled 1).  All ports and Ethernet are 1 Gbps, while SL downlink typically maxes at 200 Mbps, usually less.  All switches are Omada.    Note the Y column numbers.  

 

​  ​

Outcome is to avoid blocking on one WAN while the other has lots of capacity left.   The result seems the same (unbalance) whether the system is heavily or lightly loaded.  Majority of traffic goes to one WAN port.    

 

I've tried policy rules to split traffic as a load balance method, but it becomes dependent on which client group is drawing traffic on any particular day and disallows overflow to use the other WAN connection.    

 

Would artificially skewing the declared balance ratio to counteract the 1 to 5 balance that the router seems to be deciding. Like entering 1:5 ?    But now I'm second guessing the load balance code which may worsen results.  

 

"How load balancing works to optimally distribute traffic?  Why it appears so unbalanced?"

 

Thanks!  

If you have two WAN active, for things like Starlink, it might have a constant connection. It might not end or become too continuous that the conversation(session) cannot be ended.

And for your network, if a device is constantly connecting to a service, for example, if you have an IoT and it reports to the server every 10 minutes, it might not disconnect the session and will preferably go to the same WAN.

In statistics, you will see one of the WANs will have more traffic than the other.

Policy routing would make sense if you set it to the priority mode. With that, it will not lose the link backup and keep a persistent connection.

Load balancing has an algorithm that I cannot analyze or break down from the code level. I don't think I even could do it due to the confidentiality. But it basically works in the guide I wrote if you happen to read it.

Common Questions About the Load Balancing, Link Backup(Failover) & Online Detection