IBM United States Software Announcement
210-008
IBM is a registered trademark of International Business Machines Corporation
21
expected to markedly reduce dump capture time when a large amount of data
must be paged in during SVC dump processing. Internal IBM laboratory tests have
shown that the capture time for SVC dumps can be reduced by over 60% for large
dumps with a substantial percentage of data on auxiliary storage. The amount of
improvement is expected to vary depending on how much data must be paged in
during dump processing, how much real storage is available to the system, and
the system's workload.
• Standalone Dump will be designed to better prioritize data capture for address
spaces, and to dump a number of system address spaces first irrespective of their
ASID numbers. This is intended to attempt to capture the data most often needed
to diagnose system problems more quickly in case there is not enough time to
take a complete standalone dump. Also, Standalone Dump will be designed to
allow you to specify additional address spaces to be added to the predefined list
using a new ADDSUMM option.
• Dump processing will be designed to act on a new option for the CHNGDUMP and
DUMP commands, and in parmlib member IEADMCxx. The new DEFERTND option
will allow you to specify that task nondispatchability for address spaces being
dumped as a result of a DUMP operator command be delayed until after global
data capture is complete. This is intended to reduce the amount of time tasks and
address spaces being dumped are set nondispatchable to capture volatile data to
reduce the impact of command-initiated SVC dumps.
• The existing XCF/XES CF structure hang detect support is planned to be extended
by providing a new CFSTRHANGTIME SFM Policy option that will allow you to
specify how long CF structure connectors may have outstanding responses. When
the time is exceeded, SFM will be designed to drive corrective actions to try
to resolve the hang condition. This is intended to help you avoid sysplex-wide
problems that can result from an affected CF structure that is waiting for timely
responses from CF structure connectors.
• One focus area in z/OS V1.12 is the time it takes to shut down and restart the z/
OS system itself and major subsystems such as DB2. Substantial reductions in
shutdown and restart times for DB2 systems that use a large number of data sets
are expected in addition to improvements in the time required for some phases of
z/OS initialization processing. Planned improvements include:
– Design changes for Allocation intended to improve performance for address
spaces that allocate a large number of data sets in a short time. These changes
are expected to markedly reduce the startup time for these address spaces,
such as DB2 address spaces and batch jobs that process a large number of data
sets per job step.
– Changing subsystem initialization from serial to parallel for initialization routines
that are listed in IEFSSNxx parmlib members after a new BEGINPARALLEL
keyword, to allow you to reduce system startup time by allowing many of these
routines to run in parallel.
– An XCF design change to help reduce IPL time when very large sysplex couple
data sets are in use.
• z/OS Communications Server plans to introduce sysplex distributor support for
a hot-standby server through the use of a new distribution method, HotStandby.
You configure a preferred server and one or more hot-standby servers. The
preferred server with an active listener receives all new incoming connection
requests, and the hot-standby servers act as backup servers should the
designated preferred server become unavailable. The hot-standby servers can be
ranked to control which hot-standby server becomes the active server. You can
also control whether the sysplex distributor automatically switches back to using
the preferred server if it again becomes available, and whether the distributor
automatically switches servers if the active target is not healthy.
• z/OS Communications Server sysplex problem detection and recovery is planned
to be enhanced to detect when the TCP/IP stack has abended five times in less
than a minute. Existing sysplex recovery logic is applied when this problem is
detected.
Optimization and management capabilities
With the ability to intelligently manage workloads, reprioritize work, dynamically
reallocate system resources between applications quickly and efficiently, and help
