With z/OS® V1R8, Parallel Sysplex® environments can realize significant improvements in availability especially during recovery processing for coupling facility resources. The message-based protocol offers advantages over the existing policy-based protocol for Parallel Sysplex configurations that have a large number of structures and structure connectors, especially during scenarios involving system failure and coupling facility recovery.

For example, consider the role of the CFRM couple data set in a recovery situation. With the policy-based mode of CFRM processing that exists prior to z/OS V1R8, the CFRM couple data set, as the centralized repository for coupling facility resource usage, often becomes a bottleneck when recovery for a failed system or coupling facility is required. This recovery implies recovery for multiple structures and connectors. Each element of recovery requires every surviving system to access the CFRM couple data set under serialization, implying that access is limited to one system at a time. Because each system potentially must access the CFRM couple data set many times, contention on the data set greatly slows the recovery process.

In contrast to the policy-based protocol, the new message-based protocol available with z/OS V1R8 defines a single system as "manager" and all other systems as participants in the recovery process. The manager system is responsible for coordinating events and confirmations with the participant systems and is responsible also for updating the CFRM active policy, thus reducing I/O to the CFRM couple data set to a single, central control point. Communication between the manager system and the participant systems, on which applications or subsystems are running, is through XCF signaling.

Message-based processing applies on a structure basis. XCF signaling structures are not eligible to use message-based processing and will continue to use the policy-based protocol.

With z/OS V2R1, message-based processing is enhanced with the ability to perform CFRM LOSSCONN recovery management. When the optional CFLCRMGMT function is enabled on all active systems in the sysplex, the message-based manager system may enable CFRM LOSSCONN recovery management. CFRM LOSSCONN recovery management may enhance average CF LOSSCONN recovery time by processing CF structures serially, rather than in parallel. Without CFRM LOSSCONN recovery management, systems independently initiate processing (rebuild, for example) to recover from a loss of connectivity to a CF and all that processing is initiated simultaneously. Performing recovery processing for a large number of structures in parallel tends to result in recovery processing for each structure taking a long time. With CFRM LOSSCONN recovery management, the message-based manager system is responsible for initiating processing to recover from a loss of connectivity to a CF and all that processing is initiated in a more serial manner. The serial nature of the recovery processing may help reduce interference and reduce average recovery time.

To enable message-based processing, a new version of the CFRM couple data set is required. The IXCL1DSU format utility supports the new keyword ITEM NAME(MSGBASED) to indicate that the CFRM couple data set is to be formatted to support message-based processing. Once a primary CFRM couple data set that supports message-based processing is active in the sysplex (either at system initialization time or through a pswitch when the sysplex is switching to a new CFRM couple data set), message-based processing can become enabled for each structure for which there are no events pending delivery for either the structure or connectors to the structure. See CFRM parameters for format utility for information about formatting the CFRM couple data set.

IBM® recommends that an installation enable the new message-based processing protocol because of the large performance, availability, and scalability benefits that message-based processing provides. Once all systems are at z/OS V1R8 and there is no intention of falling back to a lower level of z/OS, message-based processing can be enabled and also non-disruptively, disabled, according to the installation's requirements. The installation can also change the manager system while message-based processing is enabled without affecting the sysplex operation.