Per packet RR LB for next hop

Hi,

The LINK::nexthop command is read only you can not interact with it: Click here

To load balance each packet you can try something like this:

 
 when CLIENT_ACCEPTED { 
 set first_bind 1 
 TCP::collect 
 } 
  
 when CLIENT_DATA { 
 if { $first_bind } { 
 set first_bind 0 
 } else { 
         LB::detach 
        } 
       TCP::release 
       TCP::collect 
 } 
  
 

You should activate oneconnect at the same time.

HTH

it wasn't LINK::nexthop i was referring to, but the global nexthop command - http://devcentral.f5.com/wiki/default.aspx/iRules/nexthop.html

 

 

I think I can see the logic in that, lets have a bash at it. thanks.

 

 

Ahh I tried, the irule requires a TCP profile, but this is to apply to an ip forwarding virtual, so can't have a TCP specific profile.

any clues guys?

is this tcp traffic? You're better off balancing by stream rather than packet by packet for TCP... That way the packets arrive in order and the endpoint doesn't have to buffer & process out of order tcp packets which can be detrimental. This is the main reason CISCO routers/switches will (By default) put a stream over one link of an etherchannel rather than separating the load packet by packet.

 

 

It'll also be more economical on the F5 performance, since packet by packet won't use the ASIC (As you'll have na iRule on it).

 

 

H

Personally I'm in favour of stream based ones, but it depends how the traffic really looks as to what would work best. We have a gig link with 10 x 100mb tunnels across it. So *IF* there is traffic which needs to push above 100mbps then if it's stream based that's never going to happen, and as is often the way, a document with a bunch of signatures on it states per packet.

 

 

I can't see how per packet would be better, and i do have it going really well with standard observed lb, but ideally need at least a proof of concept if nothing else.

 

 

btw, is there any information on how / if the load balancing methods change when a pool is being used for routing as opposed to a vip? things like observed and predictive methods don't feel as applicable.