Google
 

Trailing-Edge - PDP-10 Archives - tops10_tools_bb-fp64b-sb - 10,7/klbpa/klbpa.man
There are 4 other files named klbpa.man in the archive. Click here to see a list.



                       KLBPA %1(1) users reference manual




















Copyright (C) 1979
Digital Equipment Corporation, Maynard, Massachusetts, U.S.A.

This software is furnished under a license and may be used and  copyed  only  in
accordance  with  the  terms of such license and with the inclusion of the above
copyright notice.  This software or any other copies thereof may not be provided
or  otherwise  made available to any other person.  No title to and ownership of
the software is hereby transferred.

The information in this software is subject to change without notice and  should
not be construed as a commitment by Digital Equipment Corporation.

Digital assumes no responsibility for the use or reliability of its software  on
equipment which is not supplied by Digital.
KLBPA %1(1) Users Reference Manual                                        Page 1
Introduction


     KLBPA (KL-10 Background Performance Analysis) is a program to  control  the
gathering  of  general  "background" system performance statistics of a 7-series
TOPS-10 monitor via the performance meter available on all KL-10 cpus.  As such,
KLBPA is meaningful only for KL-10 based TOPS-10 systems.

     In order to use KLBPA you must be running under the OPR account.
KLBPA %1(1) Users Reference Manual                                        Page 2
Background Performance Analysis Overview.


     KL-10 processors come equipped with  a  very  powerful  and  very  flexible
analytical  tool  called  the  "Performance Meter" or "Performance Meter Board",
hereinafter referred to as simply the "meter".  The meter allows measurement  of
a  wide  variety  of  operating  conditions or events which take place under the
control of the KL-10 cpu.

     The meter operates in one of two modes - duration mode or event  mode.   In
duration  mode  the  meter  measures  the  amount  of  time  (the duration of) a
condition occurs.  In event mode the meter counts  how  many  times  (events)  a
condition occurs.

     The meter can measure user mode  versus  executive  mode,  interrupt  level
processing, RH20 channel to memory usage, memory usage (from which the cache hit
ratio can be calculated), and a number of other conditions.  The meter can  even
measure  a  microcode-controlled  condition, enabling the KL-10 to, for example,
count the number of times an ILDB instruction is executed (or the amount of time
spent  executing  ILDB  instructions).  To use the microcode in conjunction with
the meter requires modifying the microcode.

     For details  on  exact  meter  operation  read  the  DECsystem-10  Hardware
Reference Manual.

     The meter is available to user programs via the PERF.  monitor  call.   For
details  on  how  to  use  the PERF.  monitor call read the TOPS-10 7.00 Monitor
Calls Manual.

     Starting with the 7.00 release of the TOPS-10 monitor  the  monitor  itself
can use the meter to gather system-wide operating statistics independent of user
programs.  If a specific user program has a cpu's meter in use then the  monitor
can  not  use that cpu's meter (when the job releases the meter the monitor will
automatically resume its background usage, if enabled).

     If the meter is available to the monitor then the monitor will  (on  orders
from,  for  example,  KLBPA)  enable  the  meter  and  leave  it running "in the
background" gathering statistics.

     The monitor has a command table (in the COMMON module) which contain  meter
commands.   The  monitor  will  round-robin  through  the  command table passing
successive conditions to the meter.  A user-settable sampling interval  controls
how often the meter command is changed.

     At the end of a sample interval the monitor reads the meter value  and  the
elapsed  time  and adds them into a CDB variables subtable (GETTAB item %CCBPA),
then steps to the next command, reenables the meter, and continues.  The monitor
sampling is driven by the monitor clock interrupt service, which, running at the
APR PI level (the highest ever used by  the  monitor),  gives  a  very  accurate
sampling interval.

     It is important to note that the monitor has no  idea  what  the  meter  is
doing,  the  monitor  merely  picks  up  the 36-bit meter command from the meter
command table, passes the command to the meter, then later reads what the  meter
has  "counted",  as  well  as the interval for which the meter was enabled.  Any
interpretation of the statistics so gathered is left to user programs.
KLBPA %1(1) Users Reference Manual                                        Page 3
Background Performance Analysis Overview.


     With the 7.00 release of the  monitor  the  meter  commands  table  has  16
(decimal) entries, the first eight of which watch the RH20 to MBOX interactions,
and the second eight of  which  watch  the  amount  of  time  spent  at  various
interrupt levels.  All sixteen entries are duration mode.  

     KLBPA  itself  does  not  do  anything  with  the  background   performance
statistics,  the  ONLY  function  of  KLBPA  is  to  control whether or not said
statistics are gathered.  SYSDPY %434(562) will display these statistics as part
of  the "system" display for any cpu which is enabled to maintain them.  See the
SYSDPY manual for details.
KLBPA %1(1) Users Reference Manual                                        Page 4
Commands


     KLBPA will prompt the user with a "/" character.  At this  point  the  user
may  type  in  a  single command to KLBPA.  Commands to KLBPA are of the form of
action keywords, optionally followed by any arguments (similar to BACKUP).

     The commands to KLBPA are:

[NO]ALL         Enable [Disable] all cpus.  The "ALL" command instructs KLBPA to
                turn  on  the performance meter for all cpus in the system.  The
                NOALL command instructs KLBPA to turn off the performance  meter
                for all cpus in the system.

[NO]CLEAR       Do [not] zero the statistics tables when enabling  a  cpu.   The
                CLEAR  command  instructs  KLBPA  to  have  the monitor zero the
                per-cpu statistics table when the cpu is enabled for performance
                measurements.   The NOCLEAR command instructs KLBPA to not cause
                the per-cpu statistics  table  to  be  cleared  when  a  cpu  is
                enabled.   The  [NO]CLEAR  command is a status switch which will
                affect all subsequent enable/disable commands.

[NO]CPU0        Enable [Disable] only CPU0.  The CPU0 command enables background
                performance  statistics  gathering  only  for  CPU0.  The NOCPU0
                command disables  background  performance  statistics  gathering
                only for CPU0.

[NO]CPU1        Enable [Disable] only CPU1.  The CPU1 command enables background
                performance  statistics  gathering  only  for  CPU1.  The NOCPU1
                command disables  background  performance  statistics  gathering
                only for CPU1.

[NO]CPU2        Enable [Disable] only CPU2.  The CPU2 command enables background
                performance  statistics  gathering  only  for  CPU2.  The NOCPU2
                command disables  background  performance  statistics  gathering
                only for CPU2.

[NO]CPU3        Enable [Disable] only CPU3.  The CPU3 command enables background
                performance  statistics  gathering  only  for  CPU3.  The NOCPU3
                command disables  background  performance  statistics  gathering
                only for CPU3.

[NO]CPU4        Enable [Disable] only CPU4.  The CPU4 command enables background
                performance  statistics  gathering  only  for  CPU4.  The NOCPU4
                command disables  background  performance  statistics  gathering
                only for CPU4.

[NO]CPU5        Enable [Disable] only CPU5.  The CPU5 command enables background
                performance  statistics  gathering  only  for  CPU5.  The NOCPU5
                command disables  background  performance  statistics  gathering
                only for CPU5.

TICKS:nnn       Set measurement cycling interval to nnn clock ticks.  The  TICKS
                command  sets  the interval between successive samples read from
                the performance meter.  The TICKS command  is  a  status-setting
                command  which  affects  all  subsequent enabling commands.  The
                default value for the sampling interval is one clock tick.