HP 3000 Manuals HP 3000 Manuals
HP Performance Collection Software Enhanced
by Gerry Wade & Gary Robillard
Software Technology Division
HP Performance Collection Software (PCS) is the HP 3000 part of the
LaserRX/MPE product. HP PCS is also included in the GlancePlus PAK
product.
Workload Manager Workgroups Can Be Defined as Applications
A new product available with MPE iX 5.0 is the Workload Manager
(B3879AA). The Workload Manager allows workgroups to be defined in
addition to the 5 standard scheduling queues (A,B,C,D and E). To look at
processes that belong to a particular workgroup, a new parameter,
WORKGROUP, has been added to the Scope Parm file. WORKGROUP can be used
to define processes that are added to an application, and also can be
associated with a Workload Manager Workgroup. HP PCS adds all of the
resource usage defined by the Workgroup for these processes to the
application.
For example, look at the following application definitions:
Application = Pri Users
Queue = B
Workgroup = HI_PRI_USERS
Application = C Queue
Workgroup = CS_DEFAULT
The first application definition creates an application named "pri
users". Any processes in the BS queue would be added to this application
AND all of the processes that are part of the "hi_pri_users" workgroup
would also be added to the "pri users" application.
The second application definition creates an application named "C Queue".
Any processes in the CS Queue are added to that application.
The 5 original scheduling queues that have always been available with MPE
iX are named:
* AS_Default
* BS_Default
* CS_Default
* DS_Default
* ES_Default
These queues are available whether the Workload Manager product is
installed or not.
Note that a process is never counted twice, so the first application that
a process matches is where all of the resources used by the process are
accumulated.
Global Memory Fault Metrics Added
Eleven new performance metrics have been added to the Global data type to
provide more information on main memory shortages. These metrics can be
accessed using the extract program's "export" feature to copy data from
the log files into ASCII, Spreadsheet, or binary format files. To export
these new metrics, add the following metric names to your REPORT files:
TRANSIENT_FAULTS (Memory faults/second for transient
objects)
NMSTACK_FAULTS (Memory faults/second for Native Mode Stacks)
CMSTACK_FAULTS (Memory faults/second for Compatibility Mode Stacks)
NMCODE_FAULTS (Memory faults/second for Native Mode program files)
CMCODE_FAULTS (Memory faults/second for Compatibility Mode programs)
CMDATA_FAULTS (Memory faults/second for Compatibility Mode XDS)
NMLIB_FAULTS (Memory faults/second for Native Mode Libraries)
CMLIB_FAULTS (Memory faults/second for Compatibility Mode Libraries)
FILE_FAULTS (Memory faults/second for data files)
TOTAL_MEMORY_FAULTS (Memory faults/second for all object types)
MEMMGR_CLOCK_CYCLES (Number of memory manager clock cycles per hour)
POSIX Program File Names Supported
Programs that are run under the POSIX Hierarchical File System (HFS) have
their POSIX pathname in place of the MPE program filename in the process
data type. Process names are in one of the following formats:
:commandname For MPE Command Interpreters
/dir/dir/dir/filename For POSIX processes where the pathname <= 24 bytes
...dir/dir/dir/filename For POSIX processes where the pathname > 24 bytes
file.group.account For processes in the MPE traditional directories
sh.hpbin.sys For POSIX shell processes or terminated POSIX process
Applications can include POSIX files in their definitions according to
the following rules:
* The pattern used in the FILE=, INTERACTIVE=, or BATCH= parameters
must always follow the rules for MPE files (FILE.GROUP.ACCOUNT
with MPE style wildcards accepted).
* Files in traditional MPE directories are matched as they always
have been.
* Files in POSIX directories are matched by constructing an "MPE
Style" filename as follows:
* All non-alphanumeric characters are eliminated from the
path.
* All lowercase alpha are shifted to uppercase.
* The first 8 characters of the filename are used as a
filename.
* The first 8 characters of the first directory are used as
the account.
* The first 8 characters of the second directory are used as
the group.
For example: The POSIX file
/usr/performance/bin/utility.and.such would translate to
UTILITYA.PERFORMA.USR.
Command Line Interface for Extract and Utility
The extract and utility programs allow commands to be entered through the
run-time ;INFO= string. Multiple commands may be entered separated by a
semicolon. For example:
:RUN UTILITY.SCOPE.SYS;INFO="SCOPE KILL"
:RUN EXTRACT.SCOPE.SYS;INFO="PROCESS KILLED;REPORT REPTFILE;EXPORT DAY-1"
Commands may also be entered in the shorter UNIX format:
UTILITY.SCOPE.SYS "-sk"
EXTRACT.SCOPE.SYS "-k -r REPTFILE -xp DAY-1"
See online help for full details.
User Selectable Summarization Interval for Exported Data
The default interval for exporting global and application summary data is
60 minutes. This default may now be changed using the SUMMARY= keyword
in your report file.
SUMMARY=15 Selects 15 minute summaries
SUMMARY=1440 Selects 1440 minute (1 day) summaries
User Selectable Missing Data Value for Exported Data
When exporting data from the log files, numeric data items that are
unavailable are normally written as zero. Unavailable metrics may be
seen when exporting application data in the multiple layout format and
when an application was not active during an interval. Global memory
fault metrics are missing on log files collected by any version of
SCOPEXL prior to version B.nn.05. If you want to differentiate between
missing data and data with a value of zero, you may change the missing
value. Add a line similar to the following line to your report file.
MISSING=-99.99
Start and Stop Dates Can Be Set Relative to Log File Contents
The extract and utility programs can set the starting and stopping dates
based on the first and last date in an input log file. For example:
START FIRST Starts at the first date in the log file
STOP FIRST+7 Stops 7 days after the first date in the log file
STOP LAST Stops at the last date in the log file
START LAST-10 Starts 10 days before the last date in the log file