External IP workload balancing solutions

IBM®'s Network Dispatcher (part of WebSphere® Edge Server) and Cisco's Content Switching Module (CSM) are examples of external IP workload balancing solutions. Such solutions exist outside the sysplex, but can direct work into the sysplex. External IP workload balancing solutions typically define a single IP address representing all instances of the server, and then balance new work requests (for example, new TCP connection requests) among available servers.

External load balancers typically use a cluster IP address to represent the set of applications being load balanced. Client applications use this cluster IP address as the destination IP address for all their requests. External load balancers might be capable of using different methods for forwarding packets to their destinations. Various load balancing solutions might use different terms when describing these methods. In this discussion, dispatch mode and directed mode are discussed as two examples of these methods.

When a load balancer uses dispatch mode, the destination IP addresses for incoming IP packets is not changed. Instead, the load balancer forwards the packet to a target z/OS® system by using the MAC address of a network adapter on that system. This technique works well when the target systems and the load balancer are attached to the same network (that is, there are no intervening routers). The receiving z/OS system inspects the destination IP address of the packet, and if it matches one of the IP addresses in its HOME list, it accepts the packet. As a result, with dispatch mode, all target z/OS systems must have the load balancer's cluster IP address defined in their HOME list. It is important, however, that these addresses are not advertised externally through dynamic routing protocols. One way to accomplish this is by defining these IP addresses as loopback addresses on z/OS.

Dispatch mode also has special considerations when the load balancer is more than one hop away from the target systems (that is, a packet must be routed to the target), or when multiple z/OS target systems share the same OSA-Express® adapter. In these scenarios, you can use the following techniques to route packets to the correct target system:

On the other hand, when a load balancer uses directed mode, the load balancer converts the destination IP address (that is, the cluster IP address) to an IP address owned by the target z/OS system, using technologies such as network address translation (NAT). When IP packets for these connections are sent back to clients, the load balancer converts the source IP address (that is, the target z/OS system's IP address) back to the cluster IP address that the application had used on its request.

Some load balancer configurations might also perform NAT on the client's IP address. The client's IP address can be replaced with an IP address owned by the load balancing device. This might be necessary to ensure that all outgoing packets from a target system traverse the load balancing device, so that NAT can be performed to change the server's IP address back to the cluster IP address that the client had originally used. Therefore, it is important to note that with directed mode solutions, the IP addresses of load balanced connections reaching the sysplex might not reflect the real IP addresses of the clients making the requests. This can be an important consideration if any definitions or configuration within the sysplex rely on the client's IP address being visible on incoming connections.

External load balancing solutions might provide several configuration options that influence how workload is distributed, such as round-robin, weighted round-robin, and so on. Some of the solutions might also obtain recommendations from components inside the z/OS sysplex that might affect their workload balancing decisions. For more information, see Sysplex-aware external load balancing solutions.

Parent topic: Workload balancing