In Figure 1, a set of DataPower® appliance instances (DATAP) are configured to handle a specific Web service request and to invoke a backend z/OS® application (CICS®) to complete the processing, prior to sending a response to the Web services client. Sysplex distributor is used for load balancing of both processing tiers, the DataPower tier and the z/OS application tier, both of which are clustered for high availability and scalability.
This configuration consists of two main components:
- A tier 1 distributed DVIPA is defined to represent the cluster
of DataPower appliances
that can process a client Web service request to TCP port 8080.
For each TCP connection request sent to the tier 1 DVIPA and port, the tier 1 sysplex distributor makes a load-balancing decision and routes the request to one of the eligible DataPower target instances. These targets support a TCP control connection with the tier 1 distributor that enables dynamic load balancing feedback to and from the DataPower appliances and sysplex distributor.
Periodically, the DataPower appliances send weights that are based on the overall CPU usage of the appliance. The weights enable a type of target-controlled distribution, and the distributor makes optimized load-balancing decisions based on the target appliance's overall capacity and available capacity as compared to other target DataPower appliances.
Connection state information is passed over the control connection. The connection state information is used to determine whether a target server is available, and also to determine the states of the connections (established or terminated) that are distributed to the target DataPower appliances. The distributor uses GRE to send packets to the targets. Because GRE is used, the packets from the DataPower target back to the client do not need to traverse the tier 1 distributor. The connection state information enables sysplex distributor to maintain its active connection-routing entries and is also used to enable nondisruptive takeover of existing connections by a backup tier 1 distributor.
A DataPower distributed DVIPA can specify any of the following types of distribution methods:
- Target controlled (configured as TARGCONTROLLED)
- Round-robin (configured as ROUNDROBIN)
- Weighted active (configured as WEIGHTEDACTIVE)
Guideline: Target controlled is the preferred tier 1 distribution method, because the distributor can make optimized load-balancing decisions based on the current processing availability of the targets. - Processing for the request is completed by a z/OS application tier.
After the DataPower targets complete their processing of the inbound Web service request, they establish a connection and send the work request to a z/OS application tier.
You can configure the DataPower targets to use one of their own distribution methods and to establish connections directly to the z/OS servers, or to use sysplex distributor for a more optimized load-balancing decision. In Figure 1, sysplex distributor is shown. The distributed DVIPA that represents the z/OS application tier does not require any special configuration, and you can select any distribution method that is supported for z/OS distributed DVIPAs; however, SERVERWLM provides the most granular WLM recommendations, along with visibility into the health of the z/OS application tier.
After sysplex distributor routes the request to the target CICS application, it is processed and the results of the request are sent to the DataPower instance that originated the request. DataPower can then perform any necessary outbound processing and send a response back to the client that originated the Web service request.