In a typical load balancing scenario, server responses to client requests are routed through a load balancer on their way back to the client. The load balancer examines the headers of each response and can insert a cookie before sending the server response on to the client.
In a Direct Server Return (DSR) configuration, the server receiving a client request responds directly to the client IP, bypassing the load balancer. Because the load balancer only processes incoming requests, load balancing performance is dramatically improved when using DSR in high bandwidth applications. In such applications, it is not necessary for the load balancer to receive and examine the server’s responses. So the client makes a request and the server simply streams a large amount of data to the client.