RRS uses six log streams that can be shared by multiple systems in a sysplex. If more than one system wants to use the same set of RRS log streams at the same time, then the log streams must reside in structures. Each system that wants to connect to the set of log streams will then need to have access to the coupling facility (CF) and the DASD on which the system logger offload data set will reside. If only one system with one RRS image is going to use a particular set of log streams, or the log streams are used in a sysplex in which information should not be shared among RRS images, then the log streams can be defined as DASDONLY log streams. If DASDONLY log streams are used, logger allocates a staging dataset, not a structure, to which to write data in the interim, so no CF is required. z/OS MVS Setting Up a Sysplex contains information about the tasks you need to perform related to the system logger set up. See Defining the log streams for specific details related to RRS.

The RRS images on different systems in a sysplex run independently. However, RRS images that are in the same log group share log streams to keep track of the work. If a system in the same log group fails, RRS on a different system in the same log group in the sysplex can use the shared logs to take over the failed system's work. If there is only one system connected to the structure-based log streams, or the log streams are DASDONLY, then no other system will take over the failed system's work. Any outstanding syncpoints will be resolved when RRS restarts using the logging group, and the resource managers become active within that logging group.

Table 1 summarizes the RRS logs. In the figure, gname is the log group name. A log group is a group of systems that share an RRS workload. The default log group name is the sysplex name.

Use caution when setting up logging groups because RRS is unaware of separate groups, and resource managers can restart in an unintended group. IBM® recommends using ARM to control restart locations.

To use a log group name different from the sysplex name, define the name on the procedure used to start RRS. Otherwise, the name defaults to the sysplex name. See Creating a cataloged procedure for starting RRS for more information.

To minimize any risk of losing data for structure-based log streams, you can specify that a staging dataset should be used as the duplexing medium at all times. Specifying DUPLEXMODE(UNCOND) when defining a structure-based log stream tells system logger to allocate and use staging datasets as the duplexing medium on all systems connected to the log stream. If an error occurs in the coupling facility's data, the DASD backup is a reliable copy of valid data that is available for restart.

Note that duplexing the logs can significantly slow performance, and RRS will run effectively without duplexing. But, if RRS logs are damaged, RRS might be unable to maintain integrity for the work it coordinates, resulting in inconsistent resources or RRS failure.

While your installation must decide on the risk it can afford to take, IBM strongly recommends that you use unconditional duplexing for both the resource manager data log (ATR.gname.RM.DATA) and the restart log (ATR.gname.RESTART), because any loss of data, unresolved gap, or permanent error against either of these logstreams will force an RRS cold start. The RM.DATA and RESTART logs are small and infrequently updated, so the impact on performance is minimal.