6. DATA SOURCES › 6.4 Operating System Details › 6.4.3 PR/SM and LPAR
6.4.3 PR/SM and LPAR
Processor Resource/Systems Manager (PR/SM) is a hardware
facility that enables logical partitioning of a physical
processor complex, provides additional byte-multiplexer
channel capability, and supports the VM multiple preferred
guest function.
In Logically Partitioned (LPAR) mode, the resources of a
processor complex can be distributed among multiple logical
partitions, where a logical partition is a virtual processor
complex that can run a system control program.
The performance of a logical partition is controlled by:
o The number of logical processors online. A logical
processor is the name used to describe a central processor
when in LPAR mode. A central processor is part of a
processor complex.
o Whether processors are dedicated to or shared by logical
partitions.
o The processing weight assigned to each logical partition
and whether the capped function is turned on or off.
o Whether wait completion is on or off.
o The processor dispatch time interval.
The CA MICS Analyzer Option for VM/CMS provides measurements
and status indicators describing PR/SM LPAR operations when
VM is running in at least one logical partition. The VMXLCP
(VMX Logical Processor Activity) file provides information
about each logical processor's activity in a logical
partition, while the VMXLPR (VMX Logical Partition Activity)
file provides measurements for each logical partition in the
processor complex.
PROCESSORS
Processors can be dedicated to a logical partition or shared
among multiple logical partitions.
Processors that are dedicated to logical partitions are
always available for use, but the capacity that is not used
by that partition cannot be used by other partitions.
A shared logical processor can be dispatched on any sharable
physical processor. Therefore, there is no correspondence
between the shared logical processors and the physical
processors on which the logical processors are dispatched.
For example, the following diagram shows a site that has one
processor complex containing four physical processors and
three logical partitions. One LPAR defines one logical
processor and the other two define three logical processors
apiece.
+-----------------------Processor--Complex-----------------------+
| +---------+ +---------+ +---------+ +---------+ |
| |Processor| |Processor| |Processor| |Processor| |
| | 1 | | 2 | | 3 | | 4 | |
| +---------+ +---------+ +---------+ +---------+ |
+-----------------------Processor--Complex-----------------------+
+------------------+ +------------------+ +------------------+
| LPAR X | | LPAR Y | | LPAR Z |
| | | | | |
| Logical Processor| |Logical Processors| |Logical Processors|
| ---- | | ---- ---- ---- | | ---- ---- ---- |
| |X1| | | |Y1| |Y2| |Y3| | | |Z1| |Z2| |Z3| |
| ---- | | ---- ---- ---- | | ---- ---- ---- |
|------------------| +------------------+ +------------------+
If Processor 1 is dedicated to logical processor X1, the only
work Processor 1 is permitted to do is work from logical
processor X1. The work from logical processors Y1 through Y3
and Z1 through Z3 must be performed on Processors 2, 3, and
4.
If Processor 1 is shared by the logical processors, all of
the shared logical processors are candidates to be dispatched
on Processor 1. Processor 1's priorities are established by
the processing weights and implementation of the capping and
Wait Completion functions discussed later in this section.
The following data elements identify the configuration of a
logical partition:
PRSMLPNM Logical Partition Name
PRSMLPNO Logical Partition Number
PRSMTPP Physical Processor Count
PRSMNOPP Total Sharable Physical Processors
PRSMTLPS Number of Defined Logical Partitions
LPRCPUCT Number of Logical CPUs
LPRSHRMD Intervals with Shared Processors
LPRCOLCT Intervals in Measurement Collector LPAR
LCPCPUCT Number of Logical CPUs
LCPSHRMD Intervals with Shared Processor
LCPCOLCT Intervals in Measurement Collector LPAR
PROCESSING WEIGHTS
A logical partition with dedicated processors is not affected
by processing weights.
A LPAR with shared processors uses processing weights to
specify the portion of the shared processor resources
allocated to a logical partition when there is contention for
the shared processor resources.
Processing weights can range from 1 to 999. The relative
share of processing resources available to a shared partition
is calculated by dividing the processing weight of the
partition by the total processing weight for all active,
sharing logical partitions. For example, there are three
shared logical partitions defined in the complex:
Processing Number of
LPAR Name Weight Online Processors
--------- ---------- -----------------
X 30 1
Y 70 3
Z 50 3
The dispatcher allocates shared processing resources to each
of the logical partitions as follows:
Processing
LPAR Name Share
--------- ----------
X 30/150 = 20.00%
Y 70/150 = 46.67%
Z 50/150 = 33.33%
The share of processing resources for each online logical
processor is calculated by dividing the relative share for
the logical partition by the number of online logical
processors.
LCPU
LPAR Name Share
--------- ----------
X 20.00/1 = 20.00%
Y 46.67/3 = 15.56%
Z 33.33/3 = 11.11%
A logical partition can receive more than its processing
share when there is excess processor capacity, and can
receive less than its processing share when its workload
demand is lower than the capacity specified by its weight.
The PR/SM capping function enables you to limit CPU resource
usage so a logical partition will not receive more than its
processing share. Capping is done on a per logical processor
basis.
The following data elements report processing weight and
capping status for a logical partition:
LPRPPWGT LPAR Processing Weight
LPRCAPP Intervals with Capping
LCPPPWGT LPAR Processing Weight
LCPCAPP Intervals with Capping
WAIT COMPLETION
The Wait Completion setting does not affect logical
partitions with dedicated processors.
The Wait Completion setting controls whether a shared
processor is dispatched on an event-driven basis (Wait
Completion=Off) or a time-driven basis (Wait Completion=On).
With event-driven dispatching, the physical processor is
reassigned to another logical processor when the logical
processor enters a wait state or when it reaches the end of
its dispatch time slice. With time-driven dispatching,
shared processor resources are distributed on the basis of
time intervals. Logical processors are given an entire
dispatch interval even if they have no work to do.
The operator can limit the amount of continuous physical
processor running time that a logical processor can be
dispatched. Values range from 001 through 100 milliseconds.
By default, processor running time is dynamically calculated
based on the number of physically shared processors and the
number of active logical processors.
The following elements indicate the Wait Completion status of
a logical partition or a logical processor:
LPRWAITF Intervals with Wait Completion On
LPRGTS Global Time Slice
LCPWAITF Intervals with Wait Completion On
DISPATCH TIME AND MANAGEMENT TIME
LPAR dispatch time is accumulated when a physical processor
is assigned to a logical processor within a logical
partition, whether or not the logical processor is actually
busy. LPAR dispatch time includes wait time if Wait
Completion is enabled. (See the WAIT COMPLETION discussion
earlier in this section.)
There are two types of LPAR management time: time spent
managing the physical configuration and time spent managing
the partition. To differentiate between them, the CA MICS
Analyzer Option for VM/CMS includes data from an extra
logical partition with the name "PHYSICAL" in its VMXLPR and
VMXLCP files.
Time spent managing the physical configuration is contained
in the logical partition named PHYSICAL. Time spent
managing the physical configuration is not attributable to
any of logical partitions.
Time spent managing the partition is contained in logical
partitions other than the partition named PHYSICAL.
The following data elements can be used to assist in
analyzing CPU busy and LPAR management overhead:
LPRLPDTM LPAR Dispatch Time
LPREPDTM LPAR Effective Dispatch Time
LPRLPMTM LPAR Management Time
LPRAVDTM Average LPAR Dispatch per Interval
LPRPCSSU LPAR Real CPU Utilization
LPRPCESU LPAR Effective CPU Utilization
LPRPCMTM Percent LPAR Management Overhead
LPRDURA LPAR Duration
LPRMXSSU Maximum Real CPU Utilization
LPRMNSSU Minimum Real CPU Utilization
LPRMXDTM Maximum LPAR Dispatch Time
LPRMNDTM Minimum LPAR Dispatch Time
LPRMXMTM Maximum Pct LPAR Management Overhead
LPRMNDTM Minimum Pct LPAR Management Overhead
LCPLCDTM LCPU Dispatch Time
LCPEFDTM LCPU Effective Dispatch Time
LCPLCMTM LCPU Management Time
LCPLPDTM LPAR Dispatch Time
LCPEPDTM LPAR Effective Dispatch Time
LCPLPMTM LPAR Management Time
LCPPCSSU LCPU Shared System Utilization
LCPPCSIL LCPU Dispatch Share in LPAR
LCPDURA LPAR Duration
CPU UTILIZATION
To determine the CPU time consumed by a logical partition
that runs a VM system, you need to know whether the logical
processors share physical processors or have dedicated
physical processors.
Note: For a logical partition that runs z/OS, see $PRSMCPU
in the Data Dictionary appendix.
Effective real CPU busy time measures physical processor
resources that are actually consumed by the workloads within
a logical partition.
Physical Processor Dedicated to a Logical Processor:
LPRPCVPU (LPAR Logical CPU Utilization) is 100% busy.
LPREPDTM (LPAR Effective Dispatch Time) is the time that
processor resources were allocated to the partition, but
not what was actually used by VM. Use the VMXSGP file's
SGPTOTTM element (Total CPU Time Used) for effective real
CPU time consumed by VM SCP. SGPWATTM (Total System Wait
Time) is VM's idle time, but this time cannot be used by
any of other logical partitions.
Physical Processor Shared by Logical Processors with Wait
Completion On:
LPREPDTM (LPAR Effective Dispatch Time) is the time that
processor resources were assigned to the partition
(excluding management time), but not what was actually used
by VM. With Wait Completion enabled, the physical
processor is not reassigned to another logical processor
until the dispatch time slice ends. Use the VMXSGP file's
SGPTOTTM element (Total CPU Time Used) for effective real
CPU time used by the logical partition. The difference
between LPREPDTM and SGPTOTTM in a measurement interval is
the time that VM has no work to do, but a physical
processor is still allocated to a VM partition.
Physical Processor Shared by Logical Processors with Wait
Completion Off:
Use either the VMXLPR file's LPREPDTM element (LPAR
Effective Dispatch Time) or the VMXSGP file's SGPTOTTM
element (Total CPU Time Used) for effective real CPU busy
time used by the VM system. This time does not include the
LPAR management time. Even if VM has no work to do, a
physical processor is still allocated to the VM partition
so no time is wasted.
To compute the processor complex's idle time, which is the
capacity that is neither used by the workload within a
logical partition nor used by the LPAR management functions,
do the following:
1. Add the value of effective real CPU busy time from all
the partitions except PHYSICAL at the measurement
interval.
2. Add the value of LPAR management time (LPRLPMTM) from all
the partitions including PHYSICAL at the measurement
interval.
3. Multiply the LPAR duration (LPRDURA) from a logical
partition that was active during the measurement interval
by the quotient of the total number of physical processors
(PRSMTPP) divided by the number of active LPAR intervals
(LPRLPAR) during the measurement interval.
The result is the total physical processor time available
in the processor complex.
4. Idle time is total physical processor time available minus
effective real CPU busy time from all partitions except
"PHYSICAL" minus total LPAR management time (that is,
items 3, 1, and 2 listed above.)
Idle time includes the time that a physical processor is
assigned to a logical processor, but the physical
processor is not actually busy because the logical
processor is dedicated. Or, if the physical processor is
shared, Wait Completion is set to On.
