Forum Discussion

Martin_297498's avatar
Martin_297498
Icon for Nimbostratus rankNimbostratus
Apr 15, 2017

Outbound Link Redundancy

Hi Everyone

 

I have a VS for outbound traffic as default gateway with two ISP links in my BIG IP V.E. 12

 

The VS is configured as the following: L4 (performance) 0.0.0.0/0 Automap for source translation

 

I have One Pool with the two ISP links as members in round robin method.

 

The issue I noticed is that if let a ping running from a internal machine (to 8.8.8.8, for example) and then turn down the ISP being used my pings replies stops. I was expecting that the BIG was going to automatically redirect my traffic to the other ISP that is available.

 

If I just stop the ping command and starts it again I received the replies normally.

 

So it looks like when there's a outbound session started in BIG IP it keeps tracked to the pool member and is not committed to the available link in case of a link down event.

 

Is this the correct behavior or may I be missing something in my configuration?

 

Thanks. Mártin

 

application delivery
BIG-IP
LTM
  • Amine_Kadimi's avatar
    Amine_Kadimi
    Icon for MVP rankMVP
    Apr 15, 2017

    Do you have a monitor associated with your pool? How exactly do you turn down the link?

     

  • Martin_297498's avatar
    Martin_297498
    Icon for Nimbostratus rankNimbostratus
    Apr 15, 2017

    Hi Amine, thank your for replying.

     

    Yes, I do have a health monitor on the Pool, I'm using gateway_icmp. I'm also using gateway_icmp on each of the members.

     

    The links are going down and up as they should, do you believe this my be associate to the kind of health monitor I'm using?

     

    Thanks.

     

  • Soda_Cup_148395's avatar
    Soda_Cup_148395
    Icon for Nimbostratus rankNimbostratus
    Apr 17, 2017

    The ping is being treated as a connection so it is getting load balanced ONCE, which is expected behavior. When you stop the ping and try again(after simulating a failure) the ltm will NEVER load balance to a failed pool member.

     

    So you ping test is probably not the best on for testing this. Put the action on service down in the pool to 'reject' and test again using a browser or whatever application your expect to traverse the load balancer for your users.

     

    • Martin_297498's avatar
      Martin_297498
      Icon for Nimbostratus rankNimbostratus
      Apr 17, 2017

      Hi Soda Cup, thank you for your words.

       

      I was imagining that maybe the ping could not be a good test. But its odd since the users are expecting something transparent when a link outage happens.

       

      Great tip about the action on service down, I'm quit new to the BIG IP world, so I was checking that there's also the reselect option instead of reject (The BIG-IP system manages established client connections by moving them to an alternate pool member without a connection teardown or setup.).

       

      I'll test both and let you know.

       

      Thanks.

       

    • Soda_Cup_148395's avatar
      Soda_Cup_148395
      Icon for Nimbostratus rankNimbostratus
      Apr 17, 2017

      Great- how transparent a failed link could be I think depends on the service. HTTP would be transparent, clients respond well to just trying again after getting a reset.

       

      But I am not sure about something like a file transfer, I bet that would have to be retried as a new connection.