Sysplex distribution with DataPower

You can use sysplex distributor to balance workload across a cluster of non-z/OS target hosts that are explicitly enabled for sysplex distributor load balancing. An IBM® WebSphere® DataPower® appliance is an example of a non-z/OS target that supports sysplex distributor load balancing.

IBM WebSphere DataPower appliances are often used as a front-end processing tier to z/OS®, enabling and enhancing the participation of z/OS applications in a service-oriented architecture (SOA). DataPower can provide for the transparent Web services enablement of z/OS applications, and can enhance distribution of workloads that are already enabled for Web services on z/OS by providing acceleration and more efficient handling of Web services security protocols, XML schema validation, and numerous other functions.

Tips:

Although a Web services workload is described here, you can apply the concepts in this information to other protocols that DataPower supports, as long as those protocols are TCP-connection oriented.
Sysplex distribution with DataPower is available for both IPv4 and IPv6. The following examples describe IPv4 distribution with DataPower, but this function applies to IPv6 as well.

In many environments, multiple DataPower instances are placed in a cluster to provide higher availability and scalability. This requires a load balancing component that distributes incoming Web services connection requests to the DataPower appliance cluster, as shown by the first tier of load balancing in Figure 1. After DataPower has performed its portion of the processing for the incoming request, it typically routes the request to the z/OS application tier, which includes applications such as CICS®, IMS™, WebSphere, and DB2®. In a high availability sysplex configuration, the z/OS application tier probably requires a second tier of load balancing that balances requests across the z/OS application tier.

Figure 1. DataPower load balancing overview

Two configuration options for load balancing of z/OS workloads with DataPower

Figure 1 shows two options for load balancing of z/OS workloads with DataPower.

In configuration A, an external network-based load balancer is used to balance incoming Web services connection requests to DataPower. The external load balancer is configured to receive TCP connection requests for a specific port and an IP address that represents the cluster of DataPower appliances that can handle the connection request. The external load balancer selects a target DataPower appliance to route the request to based on the distribution method configured for the load balancer (round robin, weighted round-robin, and so on). When the request is processed by the DataPower appliance, a secondary TCP connection might be initiated to route a subsequent request to the z/OS application tier, which requires a second load balancing tier. You can implement this second load balancing tier using load balancing support that is built into DataPower, which by default uses a round-robin distribution, or you can use the z/OS sysplex distributor for this load balancing tier. Because sysplex distributor is an integral part of the sysplex environment, it has in-depth, real-time knowledge of the z/OS environment, including z/OS Workload Manager recommendations about current capacity and performance of each application and LPAR instance, and in-depth information about the current state and health of the z/OS application tier. This information enables sysplex distributor to make optimal load balancing decisions based on the current state of the z/OS application tier and other sysplex resources.

Configuration B is a variation of configuration A, with the main difference being that sysplex distributor is used as the load balancing component for both distribution tiers. This enables you to use a single load balancing solution for the composite z/OS workload, including both the DataPower and z/OS application processing tiers, which simplifies load balancing administration. Using sysplex distributor as the tier 1 load balancer for DataPower takes advantage of DataPower support to communicate with sysplex distributor over an out-of-band TCP connection to provide dynamic load balancing feedback and enable the following optimizations:

Routing is optimized so that outbound traffic (from the tier 1 DataPower target server towards the client) does not need to traverse the sysplex distributor.
Sysplex distributor uses generic routing encapsulation (GRE) to forward inbound distributed packets to DataPower, without performing network address translation (NAT) to map the destination IP address of the connection request to the IP address of the target DataPower appliance. Outbound tier 1 traffic does not need to traverse the sysplex distributor node and can flow directly to the client, because no reverse NAT processing is necessary.
If a planned or unplanned outage of a primary sysplex distributor instance occurs, connection information provided by the DataPower appliances enables nondisruptive tier 1 takeover of existing connections between clients and DataPower targets.
If a sysplex distributor takeover occurs, sysplex distributor dynamically discovers the state of any existing tier 1 connections from each DataPower appliance. The connection state information also enables sysplex distributor to maintain its active, distributed connection-routing entries without needing to inspect packets to determine when connections are terminated.
CPU usage information provided by the DataPower appliance enables sysplex distributor to optimize its load balancing decisions, which avoids overloaded DataPower appliances by directing new requests to DataPower instances with less usage.

For a more detailed explanation of using sysplex distributor for load balancing for composite z/OS and DataPower workloads, see the following topics: