Well, we only intend to send back the response from the real destination. Also, the client will wait for one response only. Once it receives it, it will simply close the connection and initiate a new transaction. Maybe this will terminate the session so quickly that we couldn't even forward the request to another server.
We are aware of the latency introduced to the process. But since the client is only expecting one response (and the Big IP is supposed to send only one response back to the client), the latency might not be visible to the client, I think. But, all these are just what we could image. They don't seem to be archiveable.
Our goal is provide independent audit on the incoming transction at the Big IP level. Sync the data between servers is not really an option.
Thanks,
Alan