Used together with Application Clustering, transparent scalable modes can deliver linear scalability using multiple CPU cores.
This uses RSS to distribute traffic, both upstream and downstream of the proxy application, mapping streams to the correct Onload stack. When each CPU core is associated exclusively with a single clustered stack there can be no contention between stacks.
For this use-case to function correctly, the proxy application will use the downstream client address:port on the upstream (to server) side of the TCP connection. In this way RSS and hardware filters ensure that client side and server side are handled by the same worker thread and traffic is directed to the correct stack.
In this scenario the client thinks it communicates directly with the server, and the server thinks it communicates directly with the client - the transparent proxy server is ‘transparent’.