Forum Discussion

boneyard's avatar
Aug 01, 2012

which virtual server type?

just wondering, how do you usually choose your virtual server type? i understand that in many cases it is quite clear. if you want to do l7 things you need a standard server. if you just need to pass traffic you use a l3 forwarding server.

 

 

but what in cases when you have multiple options, for example smtp, do you start with standard and change when issues occur or go straight for l4 as you dont need the l7 stuff anyway.

 

 

what is your approach?

 

4 Replies

  • Hi Boneyard,

     

     

    let me try to answer that question. The main difference between a fastl4 and a L7 virtual server becomes clear, when you look where the TCP connection gets terminated. FastL4 changes the destination ip and maybe port, maybe src ip, but the TCP connection still spans from the client to the server. This works fine, but with doing that you miss out on the TCPExpress optimisation F5 offers.

     

    So, lets assume your server is an old solaris stack, not supporting window scaling or Selective Acknowledgements. This is fine for a controlled environment, like the local lan, but when serving clients out in the wild on a high latency WAN link, there is better stuff to do.

     

     

    Therefore i use, if i can, always the L7 virtual server, which will maintain a server-side TCP connection to the old server, and will handle the client-side connection using our optimized, state of the art TCP stack, and therefore ensure best data delivery and user experiences, even on a lossy WAN link.

     

     

    So, to answer your question here, in your case one would also use the L7 virtual, in order to get the most out of your f5 devices.

     

     

    hope that helps,

     

     

    Christian

     

  • Hi Boneyard,

     

     

    let me try to answer that question. The main difference between a fastl4 and a L7 virtual server becomes clear, when you look where the TCP connection gets terminated. FastL4 changes the destination ip and maybe port, maybe src ip, but the TCP connection still spans from the client to the server. This works fine, but with doing that you miss out on the TCPExpress optimisation F5 offers.

     

    So, lets assume your server is an old solaris stack, not supporting window scaling or Selective Acknowledgements. This is fine for a controlled environment, like the local lan, but when serving clients out in the wild on a high latency WAN link, there is better stuff to do.

     

     

    Therefore i use, if i can, always the L7 virtual server, which will maintain a server-side TCP connection to the old server, and will handle the client-side connection using our optimized, state of the art TCP stack, and therefore ensure best data delivery and user experiences, even on a lossy WAN link.

     

     

    So, to answer your question here, in your case one would also use the L7 virtual, in order to get the most out of your f5 devices.

     

     

    hope that helps,

     

     

    Christian

     

  • I like Christian's logic. However, if you're not able or interested in doing any optimizations above layer 4 on the traffic to a particular virtual server, it's more efficient and faster to use a FastL4 virtual server.

     

     

    Aaron
  • thanks both, the reasoning is quite clear. usually l7, but if already known it doesnt work correctly or when you dont need it go for l4.