Figure 8 shows the operation of the cluster. For simplicity, we assume here that clients directly contact a distributor (for instance, via DNS round-robin). The figure shows the user-level processes and protocol stacks at the client and the distributor and server nodes. The client application (e.g., Web browser) is unchanged and runs on an unmodified standard operating system.
The figure illustrates the cluster operations from the time the client sends the request, to the instant when it receives a response from the cluster. (1) the client process (e.g., Netscape) uses the TCP/IP protocol to connect to the chosen distributor, (2) the distributor component accepts the connection and parses the client request, (3) the distributor contacts the dispatcher (not shown in the figure) for the assignment of the request to a server, (4) the distributor hands off the connection using the TCP handoff protocol to the server chosen by the dispatcher, (5) the server takes over the connection using its handoff protocol, (6) the server application at the server node accepts the created connection, and (7) the server sends the response directly to the client. Any TCP acknowledgments sent by the client to the distributor are forwarded to the server node using a forwarding module at the distributor node.
The operation in Figure 8 assumes that the server component chosen by the dispatcher resides on a different node than the distributor. If the server is on the same node as the distributor, then the handoff is accomplished efficiently using procedure calls in the kernel.
The operation of the cluster with a layer-4 switch is similar to that in Figure 8. However, there are some important differences. First, the choice of the distributor is made by the switch, using the WRR strategy. Second, after a connection is handed off by the distributor, the switch is notified and the subsequent forwarding of TCP acknowledgments to the corresponding server node is handled by the switch.
