White Paper on Dynamic Processor Deallocation and Dynamic Processor Resilience

NOTE: Starting HWE 0206 release of Diagnostics, the CPU monitor will keep
track of each of these types of LPMCs rather than treating them as one
type as in earlier version of the monitor.
Floating-Point Errors
Besides monitoring the Cache errors on the processors, the monitor will run tests on
the Floating-Point registers to see if they are functioning properly. These tests are run
at each Poll Interval (default 1 hr) and if the tests fail on any processor, the monitor
will take the Dynamic Processor Resilience (DPR) action – as described later - on it
right away. This action is similar to that taken in case of Cache errors. This
functionality was added to help isolate CPU failures due to errors in Floating-Point
Registers. These tests were available in the CPU Exerciser tool, which does not get
executed regularly on a system. Since the monitor is always being executed, it was an
ideal solution. The number of test-vectors to be run at each Poll Interval is user-
configurable (via a configuration verb in lpmc_em.cfg file)
Technology Overview
Dynamic Processor Deactivation (DPD)
Incorporated into HP’s 11.0 and future versions of HP-UX is the capability to take a
processor out of service while the system is running without interruption to
applications. This technology is referred to as Dynamic Processor Deactivation.
Once a processor is deactivated, the HP-UX operating system will migrate all
processes that are currently scheduled on that processor to other active processors. HP-
UX will then no longer schedule processes to run on the deactivated processor until it
is reactivated. Note that if the processor has been assigned to handle interrupts for any
I/O drivers, it will continue to do so while it is inactive.
In order for Dynamic Processor Deactivation (DPD) to be available, the system must
have at least two active processors. Dynamic Processor Deactivation is not available
for the monarch processor--the processor upon which the HP-UX kernel is running. A
processor may be deallocated programmatically via an HP-UX system call, via the
Support Tools Manager (STM) CPU Expert tool (licensed for use by authorized
service personnel).