At front node, using our application each new TCP connection from clients for HTTP request can be scheduled on desired server. Similar to DNS, we have implemented three policies for server selection at front node : random, round robin, weighted round robin (based on current load of servers). By modifying a method called select_server which takes client IP address as input and returns server IP address to which this new connection should be forwarded, scheduling policies can be easily changed. Currently, distributed IP packet rewriting mechanism is used, so only client IP address and TCP port number of client side can be used to determine which server to select.
All available information about servers and client request rate information (for predefined time interval in past and average) is accessible easily using their IP addresses. Information about all servers can also be obtained sequentially.