Sharing of history files across a sysplex

Fault Analyzer uses XCF messaging from IDIS subsystems to efficiently share PDSE history files across all MVS images in a sysplex.

Combining this method with the PDSESHARING(EXTENDED) option in the IGDSMSxx member of SYS1.PARMLIB allows concurrent writing and reading of individual fault entries from all MVS images in a sysplex without contention.

Sharing of PDS history files across a sysplex can cause unwanted contention if high activity to the same history file occurs concurrently from multiple MVS images, because the access method requires that the entire data set is enqueued upon for any updates. This problem does not apply to PDSE data sets, which only require serialization at the member level and are therefore a better choice.

Some environments have previously been set up with the unusual condition of not sharing master catalogs (or a user catalog) between all MVS images in a sysplex, and thus permitting individual MVS images to use the same data set name cataloged to a different DASD volume. Do not use this setup with Fault Analyzer history files, as the common data set name causes unnecessary contention through the IDIS subsystem and ENQ mechanisms between the MVS images. If different history files are desired between MVS images, then the use of the &SYSCLONE. substitution variable in the history file data set name should be considered as the best alternative, since the unique naming then removes the common name ENQ contention and provides a more logical and practical management.

An alternative to using shared history files is to use the Fault Analyzer ISPF interface "View" facility, which allows multiple history files to be viewed simultaneously. For details of this facility, see Setting up views.

If sharing history files between multiple MVS images, the following must be observed:
  • Systems sharing a PDSE must be members of the same sysplex and it is recommended that all systems are running with PDSESHARING(EXTENDED) in the IGDSMSxx member of SYS1.PARMLIB. (Note that the first sysplex member to IPL determines the sharing mode used and forces all subsequent members that join the sysplex to run in that mode, whether EXTENDED or NORMAL.)
  • XCF must be active.
  • GRS (or a functionally equivalent subsystem) must be running. PDSE serialization is managed using resource major names SYSZIGW0 and SYSZIGW1. There is no need to make changes to GRS RNL lists, except if a serialization product other than GRS is used.

It is possible for OPEN errors to occur against PDSE data sets that are shared across a sysplex, irrespective of the above considerations. This situation might happen as a result of a cross-system sharing conflict, typically resulting in message IEC143I RC 213-70.

For more information about sharing PDSE data sets in a sysplex, see chapter PDSE Sharing and serialization in the Redbooks® publication Partitioned Data Set Extended Usage Guide (a copy can be downloaded from www.redbooks.ibm.com) and the Sharing PDSEs section in chapter 28 Processing a PDSE in the DFSMS: Using Data Sets manual.

If an installation is required to use PDSESHARING(NORMAL) then the IDIS subsystem is also required as this provides history file access management aimed at reducing cross-system sharing conflicts. However, if a history file is accessed outside of Fault Analyzer, then message IEC143I RC 213-70 might still occur.

No special Fault Analyzer options are required to use sysplex-shared fault history files.

Tip: Make the setting of the IDIS subsystem PARM='UPDINDEX' option, or the absence thereof, the same for all IDIS subsystems in the sysplex. For details about this option, see Caching of history file $$INDEX data.