Forum Discussion

Gino_105072's avatar
Gino_105072
Icon for Nimbostratus rankNimbostratus
Jul 02, 2010

Persistence Fun & Games

Hi,

 

 

Using an LTM1500 with 9.3.0 and having a little fun with session stickiness.

 

 

Got two servers behind the LTM (oracle, something or other, basically web servers)

 

 

They are configured for persistency with cookie first, source-addr as a fallback and basic round-robin

 

 

It looke like 98% plus of connections are going to a single pool member and I can figure out why.

 

 

This morning I browsed to it and decoded the cookie and saw my pool member. I closed the browser cleared my history, cookies etc (I can see the cookie go when I close anyway). I went back and gantastic I decoded the cookie and saw I had gone to th other member.

 

 

Tried that a few times and I keep going back to the second member, regardless of what I do cookie clearing etc (even though it dies when I close the browser)

 

 

Anyone seen this kind of thing before where do I need to start troubleshooting.

 

 

Thanks,

 

 

Gino

 

2 Replies

  • Round-Robin has always been flakey in my opinion. It will count any connection attempt as that servers turn in the rotation being taken and rotate to the next server.

     

     

    I would suggest going to Least Connections (member) or Least Connections (node) and see if your connections Level out if Server Load is your major concern.

     

     

    Least Connections (member) - Load Balances per member of the pool. I doesn't care about the actual load on the server, or the number of overall connections to the server, just the number of connections balanced out in the specific pool you apply this to.

     

    Least Connections (node) - Keeps track of the total number of connections on the entire Load Balancer and compares that number to the total number of connections for the other servers in the specified pool. Server with the least number of connections gets the next connection. (with this setting it can APPEAR to be unlevel, but if you look at the comparison of the Nodes, you will see that it is even).
  • Hi Michael,

     

     

    Thanks for the reply. I will give that a go and see what results I get.

     

     

    Regards,

     

     

    Gino