Forum Discussion

Tim_Brown's avatar
Tim_Brown
Icon for Nimbostratus rankNimbostratus
Jul 24, 2009

LTM bypass after client connect

Is there a way configure a virtual on an LTM (on a stick) to simply pass a client connect to the selected node in a pool and then step out of the connection so the server can respond directly to a client?

 

 

We would like to load balance a set of remote MS Terminal Servers for the clients in that site but not force the connection through the LTM for the duration of the session.

 

 

Thanks

4 Replies

  • If it were HTTP, you could redirect the client to the actual server IP address so the client would then make requests direct to the server. I don't think there is a way to redirect a MS terminal server request. I suppose you could use nPath to allow the server to use the VIP address and respond back directly to the client. The client would continue to send packets to the VIP, but the response wouldn't go through LTM.

     

     

    That said, why are you wanting to do this? I think it would be more efficient to have LTM broker the connections. And the general idea behind terminal services is that minimal data is actually sent between the client and the server.

     

     

    Aaron
  • nPath. Interesting approach.

     

     

    It would be more efficient to utilize the LTM, except that the LTM is on the other side of the WAN, which induces additional latency to an already latency sensitive application.

     

     

    That and it's quite inefficient to hairpin connections across the WAN to access local resources.

     

     

    Tim
  • Hi Tim,

     

     

    Another option that might work would be to use DNS to handle the load balancing. This would potentially allow you to give the client the IP address of the server without establishing the TCP connection through LTM. GTM or possibly even a standard DNS server might work depending on what load balancing algorithm you want to use.

     

     

    Aaron
  • Thanks Aaron.

     

     

    Using RR A records is our second choice, but we're reluctant to use that as there's no "dead host" detection.

     

     

    Tim