Forum Discussion

aparekh's avatar
Icon for Nimbostratus rankNimbostratus
Feb 08, 2023

Uneven Load balancing Round Robin in F5

Hello Team,

We are facing uneven load balancing in one of the VIP. Simple round robin algorithm is used for load balancing. No persistence/One connect or any other attribute is configured which overcomes this roundrobin algorithm. There are 2 servers behind the VIP. What could be the possible reasons of seeing the uneven traffic on both of the backend servers.

5 Replies

  • Hi,

    So at a guess and not knowing your application you could have large queries or traffic requests coming through which is over time landing on one server rather than the other.

    If the servers are the same side, a large query might take longer to complete, so it'll slow and over time get slower and slower. (if that make sense)
    Round robin is just a 1,23,4,1,2,3,4,1 etc algorithm so not traffic driven more process driven.

    Maybe try least connections instead as a start, this looks at which servers have less connections open/sent to it.

    So its trying to look at loading more.
    the issue may come in the fact you can't see how large the queries are, but the more advanced LTM features can look at timing etc etc.

    • aparekh's avatar
      Icon for Nimbostratus rankNimbostratus

      In our environment it is not possible to change the algorithm and try to do the least connection. User's query is why the are getting more hits on first server then second. After a long investigation we found that the server's are having different CPU and memory capabilities. That's why this was happing.  

  • What is the VS and pool config (redact as needed,  generic name/ips)  config along with the show VS and pool stats?

    Are you seeing a different balance of concurrent flows over time or are the total connections to the pool members diverging over time? 

    As stated, RR is a static method that doesn't take into account any current condtions other then pool member health. 

    If pool members have changed health status, clearing the stats and observing over a time period of 100% health will get better data on this.