|In a typical setup, a WAN accelerator is placed at each end of a WAN link. The appliance sits on the LAN side, on the clear-text side of a VPN device and behind the firewall or Internet router, and intercepts all traffic. The traffic is compressed, sent across the WAN, decompressed by the remote accelerator and then forwarded to the destination.
Mike DeMaria, Breaking the WAN Bottleneck
Peter Sevcik also does a really good job explaining how WAN accelerators work and the teaches us the difference between “transparent addressing” (term used by Cisco) and “correct addressing” (term used by Riverbed).
There are two approaches used as seen by the routers between the accelerators. Transparent addressing shows the original client-server source and destination addresses and hides the addresses of the accelerators. Correct addressing shows the addresses of the accelerators and hides the addresses of client-server. Both approaches work. Both approaches have their pros and cons…
In a transparent addressing architecture, the monitoring tools will continue to show network usage by the original client-server addresses and port numbers. But you will not directly see the total traffic carried between the accelerators. Remember that they are masked.
With a correct addressing architecture, traffic monitoring tools will show traffic as having come from/to the acceleration appliances. In this case the client-server traffic volume is masked. However, in both cases the traffic volumes that are reported by routers or probes between the accelerators will be dramatically changed from the original true traffic as seen on the LANs.
In both cases the best solution is to move the network monitoring probes to the LAN-side of the appliance or to gather usage information directly from the appliance itself. Ongoing real-time before-and-after picture of what the accelerators is doing (like how much compression is being achieved) can only be supplied by the accelerators. So we recommend shifting traffic monitoring to the appliance. That way you get same accurate data from either approach.