Trailing-Edge
-
PDP-10 Archives
-
BB-FP64A-SB_1986
-
10,7/snoopy/snoopy.mem
There are 6 other files named snoopy.mem in the archive. Click here to see a list.
Program Specification
SNOOPY/TATTLE
Programs for studying
CPU time usage
Copyright 1979,1972 by:
Digital Equipment Corp.
Maynard, Mass. 01754
AUTHOR: Jud Leonard
DATE: May 31, 1972
FILE: SNOOPY.RNO
EDITION: 1
The information in this document is subject to change without notice
and should not be construed as a comitment by Digital Equipment
Corporation. Digital Equipment Corporation assumes no responsibility
for and errors that may appear in this document.
The software described in this document is furnished under a license
and may be used or copied only in accordance with the terms of such
license.
Digital Equipment Corporation assumes no responsibility for the use or
reliability of its software on equipment that is not supplied by
DIGITAL.
Copyright (C) 1972,1979 by Digital Equipment Corporation
The following are trademarks of Digital Equipment Corporation:
DIGITAL DECsystem-10 MASSBUS
DEC DECtape OMNIBUS
PDP DIBOL OS/8
DECUS EDUSYSTEM PHA
UNIBUS FLIP CHIP RSTS
COMPUTER LABS FOCAL RSX
COMTEX INDAC TYPESET-8
DDT LAB-8 TYPESET-10
DECCOMM DECsystem-20 TYPESET-11
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 2
1.0 INTRODUCTION SNOOPY is a program which samples the PC at regular
time intervals, building a histogram which describes the areas of a
program in which it spent CPU time. After sampling for a period of
time, the collected histogram is written as a disk file, for analysis
and listing by the companion program TATTLE. Options permit SNOOPY to
collect data on the monitor and/or interrupt service routines for all
jobs on the system, or to collect data on the performance of some job
or system program, including time spent in the monitor servicing the
selected job or cusp. The sampling frequency may be controlled either
by the DK10 real-time clock, or by the line-frequency clock in the
APR. The data may be collected for a single period under controlled
conditions, or SNOOPY may be set up to write histogram files
cyclically, so that data may be collected over an entire day, if
desired, without operator intervention. The resulting histogram files
are given unique extensions, and may be processed individually by
TATTLE, or combined as desired by programs written in FORTRAN or
MACRO.
1.1 REQUIREMENTS SNOOPY must lock itself in core, and must use the
RTTRP feature to work properly. It must, therefore, be run under a
PPN with LOCK and RTTRP privileges, on a monitor assembled for those
features. In addition, the TRPSET feature and privilege are required
to sample using the APR clock. While sampling may be performed from
the line-frequency APR clock, it is subject to severe bias due to the
fact that the system scheduling is also syncronized to line frequency.
Results obtained with the DK10 real-time clock are more useful, both
because of the bias problem, and because the DK10 can be made to
interrupt much more frequently, thereby increasing the number of
samples taken in a given time period, and the precision of the
results.
1.2 RESTRICTIONS SNOOPY does oot recognize dual processor (1055 or
1077) systems. It does not operate correctly when DK10 accounting is
used, or, for that matter, the DK10 is in use by any other program.
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 3
2.0 Using SNOOPY Run SNOOPY. It will ask:
SNOOP ON?(INTERRUPTS, MONITOR, JOB#, CUSP)
Answer with one of the words in parenthesis (first letter will do) and
carriage return (hereafter abbreviated <CR>). The sequence from that
point will depend on this answer, which determines the qualifiers for
entering any sample into the histogram.
2.1 INTERRUPTS This response to the first question means that the user
wants to study the time spent in the monitor during interrupt service,
and ignore any samples taken with no PI in progress. The default
histogram file name is set to INTRPT.HST. The next question is:
PI LEVEL TO WATCH?(2-7 OR *)
The user may respond with any digit between 2 and 7 inclusive, to
indicate that samples are to be taken only when the CPU is servicing
the specified level, or * to indicate that all time in PI service is
to be entered into the histogram. It is not posible to study PI level
1, because the sampling routine must be at a higher priority than the
program being studied, and 1 is the highest level. From this point
the questions and answers are the same as described in 2.11 and
following, below.
2.2 MONITOR This response to the first question means the user wants
to study the time spent in the monitor for all jobs. The default
histogram file name is set to MONITR.HST. The next question is:
INCLUDE PI TIME?(N,Y)
The user may respond with YES or NO. The default is NO, which means
that the histogram will include time spent in UUO service, but not the
time spent servicing interrupts. An answer of YES includes interrupt
service time in the histogram. From this point the questions and
answers are the same as 2.11 and following, below.
2.3 JOB# This response to SNOOPY's initial question is used to study
the performance of some particular job. The next question is:
JOB #?
to which the user should respond with the decimal number of the job he
wishes to study. SNOOPY then asks:
NAME OF CUSP?
The user should respond with the (up to) six character name of the
cusp running in the specified job. SNOOPY does not use this name
during the sampling process, but saves it in the histogram file as the
default name TATTLE should use in looking for a symbol table with
which to list the histogram data. The name is also used as the
default file name for the histogram file, with the extension .HST From
this point the questions and answers are the same as those described
in 2.7 and following, below.
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 4
2.4 CUSP This response to SNOOPY's initial question indicates that the
histogram is to reflect the time used by some specified program
whenever it is run by any job. Program is taken to mean a job with
one of the following combinations: a named low segment and a named
(possibly different) high segment or no high segment, a named low
segment and any or no high segment, or any low segment with a named
high segment. The next question is:
LOW SEG NAME?(* IF ANY)
The user may type the (up to) six character name of a low segment
which must match to include a sample in the histogram, or * to
indicate that any low segment name is acceptable. If a name is
specified, go to 2.6.
2.5 Any low segment If * is specified as the low segment name, the
next question is:
HIGH SEG NAME?
The user must specify the name of a high segment to match for samples
to be included in the histogram. Whenever a job contains the named
high segment, samples will be taken on that job. The default
histogram file name is set to the name specified, with the extension
.HST. From this point, the questions and answers are as described in
2.8 and following, below.
2.6 Named low segment If an explicit low segment name is given, it is
stored as the default histogram file name with extension .HST, and
samples are included in the histogram only for jobs whose low segment
name matches that given. The next question is:
HIGH SEGMENT NAME?(CR IF SAME)
The user may type <CR> if the high segment name to be matched is the
same as the low segment name given, * if any high segment is
acceptable, or the name of the only high segment which may be in the
job's addressing space for samples to be taken.
NOTE: A job whose low segment name matches that given, but which has
no high segment, is considered eligible for sampling, even if an
explicit high segment name was specified.
2.7 Low segment size Whenever a job number or cusp is specified for
snooping, the following questions may be asked:
LOW SEG SIZE IN K?
The user may respond with 0 or just <CR> to indicate that he is not
interested in a breakdown of time spent in the low segment (in which
case the questions continue at 2.8), or the size in K (=1024 decimal
words) of the portion of the low segment starting at relative 0 for
which he wishes to collect samples. Samples taken when the PC is
greater than the specified limit will be counted together as
out-of-range. If a non-zero size was specified for the low segment,
the next question is:
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 5
LOW SEG RESOLUTION? 2**
The user should answer with a decimal number indicating the number of
words in a group which will be counted together in the histogram.
SNOOPY shifts the low segment size right by this number to get the
size of the low segment histogram. (One additional location is
allowed for samples out of range.) The PC is divided by the resolution
when each sample is taken to determine the table location to be
incremented.
2.8 High segment size If the high segment name specified was *,
questions continue at 2.9. Otherwise, the next question is:
HIGH SEG SIZE IN K?
This question is answered in the same way as 2.7, above. If a
non-zero size is specified, the next question is:
HIGH SEG RESOLUTION? 2**
This question is analogous to the low seg resolution question in 2.7,
above, and is answered in the same way.
2.10 Monitor or UUO breakdown When the size and resolution questions
have been answered for a job or cusp, SNOOPY asks:
FULL MONITOR BRKDWN, OR BY UUO?(U,M)
This question allows the user to choose whether the time spent in the
monitor by a program is broken down by monitor addresses or by the UUO
which invoked the monitor. If UUO is chosen (the default), monitor
time spent for the job is included in a histogram by UUO code, with
CALLI broken down separately. The questions continue at 2.13.
NOTE: Due to the mechanisms by which re-scheduling can take place
during UUO processing, this option can cause UUO time to be attributed
to a job which did not perform the UUO. This time will ordinarily be
small, but the user should not be surprised to see certain UUO's
(particularly HIBER, SLEEP, DAEMON, and I/O UUO's) attributed to the
program being watched even though it has no such UUO's. If MONITOR is
chosen, monitor time spent for the job is broken down by address.
NOTE: The monitor symbol table is quite large (often 30K or more),
and TATTLE must have it entirely in core to produce a symbolic
histogram listing. Obviously, this may represent a restriction to
users with small core configurations. Therefore, TATTLE will produce
a histogram by absolute address only if it is unable to get sufficient
core for the symbol table, and the user must refer to the load map to
determine symbolic locations.
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 6
2.11 Monitor size Unless the UUO breakdown option is chosen in 2.10,
above, the following question is asked for any choice on the initial
question:
MONITOR SIZE IN K?
This question is answered in the same way as 2.7, above. If the
response is non-zero, the next question is:
MONITOR RESOLUTION? 2**
This question is analogous to the low seg resolution question in 2.7,
above, and is answered in the same way.
2.12 UUO timing If the response to the initial question was anything
but MONITOR, the dialog continues at 2.13. If monitor snooping was
selected, however, the next question is:
WATCH UUO TIMES?(N,Y)
If the user responds NO (the default), time spent in the monitor is
not broken down by UUO. If the answer is YES, however, a histogram is
built which breaks down monitor time by the UUO being serviced, with a
separate table for CALLI.
2.13 Clock to use The following question will always be asked next:
USE DK10?(N,Y)
If the user's response is NO (the default), the line-frequency APR
clock will be used to define the sampling interval, and the dialog
will continue at 2.14 (see note in 1.1). If the answer is YES, the
next question will be:
INTERVAL IN USEC'S?(200-1000000)
The user must type a decimal number between 200 and a million,
specifying the sampling interval in micro-seconds. The smaller this
number, the more samples will be taken, but the system performance
will be degraded by the time spent sampling.
2.14 Opcode breakdown The following question gives the user the option
of analyzing opcode usage, unless the sampling is done on a KI10
processor, in which case the dialog continues at 2.15:
WATCH OPCODE USAGE?(N,Y)
If the user answers NO (the default), dialog continues at 2.15. A YES
response causes a histogram of 9-bit opcodes to be built, displaying
the time spent in the watched program by each instruction.
2.15 Histogram file The next question is:
OUTPUT FILE NAME? DSK:
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 7
The user may specify a file name and/or extension to override the
default (as defined above), or may type <CR> to use the default.
2.16 Cycle time The next question is:
CYCLE TIME?(MINUTES)
The user may type 0 or just <CR> to specify that only one histogram is
to be built, or he may type a decimal number which will be interpreted
as the number of minutes during which SNOOPY will sample before it
writes a histogram file, clears the tables, and resumes sampling. The
first file written has the name specified in 2.15, above. Thereafter,
characters in the extension are forced numeric, and incremented in
decimal to 999, after which SNOOPY exits. The user may stop SNOOPY at
any time during a cycling run by typing <CR>.
2.17 Starting sampling SNOOPY will type:
DATA SIZE IS nnnn WORDS.
TYPE COMMENTS<CR> TO GO:
Where nnnn is the (decimal) size of the histogram, and represents the
minimum usable dimension of the array DATA in TATTLE. Sampling will
begin when the user types <CR>. Before that, he may enter up to 50
characters of comments, which SNOOPY will save as part of the
histogram file to be printed with the listing by TATTLE. This allows
the user to note the particular conditions under which the sampling
was done, if any.
2.18 Stopping SNOOPY SNOOPY will type:
TYPE <CR> TO STOP:
The user may type <CR> at any time to stop sampling and exit. SNOOPY
will write the histogram file to disk, and type:
HISTOGRAM FILE file.ext WRITTEN
WITH nn/mm SAMPLES.
where mm is the total number of interrupts taken, and nn is the total
number of samples included in the histogram. This message is printed
each time SNOOPY writes a file, whether in single or cycle mode.
Typing ^C's while SNOOPY is sampling will stop sampling without
writing a histogram. This is intended as an emergency abort, and
should not be used under ordinary circumstances.
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 8
2.19 SNOOPY errors The following error messages may be generated by
SNOOPY:
?TRAP LOC NOT A JSR
Attempting to setup for the APR clock, the trap location
for the APR channel was not a JSR instruction. Should
never happen.
?CANT DO TRPSET
The TRPSET UUO gave an error return, indicating either
that the user did not have TRPSET privileges, or the
monitor did not include the TRPSET feature.
?TRAP LOC CHANGED!!
Should never happen. The location on which TRPSET was
done has changed during setup.
?CANT RESTART TIMESHARING
Should never happen. SNOOPY was unable to restart time
sharing after the TRPSET UUO.
?CANT START DK10
Probably indicates there is no DK10 clock on the system.
After attempting to start it, SNOOPY found the run bit
off.
?CANT INIT DISK
When trying to write the histogram file, the INIT UUO
failed.
? ENTER FAILURE ON name.ext
The ENTER UUO failed while trying to write the named
histogram file.
SHUT DOWN
SNOOPY has trapped a monitor-detected error or ^C, and
disabled the interrupt code. SNOOPY exits to monitor
mode.
?SORRY, NEED PRIVILEGES
SNOOPY was run under a PPN without real-time trap and lock
privileges.
?x UNKNOWN
The response "x" is not recognized as valid for the
question.
?CANT GET JBTPRG
GETTAB returned an error while attempting to get the
address of the monitor table JBTPRG.
?CANT GET JBTSGN
Same as above, except JBTSGN table.
?CANT GET JBTADR
Same as above.
?CANT GET CORE
SNOOPY was unable to get sufficient core to build the
histogram tables.
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 9
?CANT LOCK
The LOCK UUO gave an error return.
?CANT GET USER IO MODE
An RTTRP UUO which attempted to get user I/O mode failed.
Probably indicates that the monitor does not have the
RTTRP feature.
? APR PRIORITY TOO LOW
The APR clock is assigned to a lower priority (higher
channel number) than the interrupt channel to be watched.
? APR CLOCK UNASSIGNED
Should never happen. The conditions read by a CONI APR,
returned a PI assignment of zero.
?CANT GET A CHANNEL
SNOOPY was unable to place itself on a appropriate PI
channel by using the RTTRP UUO.
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 10
3.0 Using TATTLE TATTLE requires a symbol table for each program
on which SNOOPY did a PC breakdown in expanded save (.XPN) format.
For the monitor, this usually exists as SYS:SYSTEM.XPN. An
appropriate file can be created for any cusp by the following
command sequence, where "cusp" represents the name of the cusp:
LOAD %S%B cusp ;INCLUDE ANY NECESSARY SUBROUTINES
SAVE cusp
R FILEX
cusp.XPN_cusp.SAV ;USE .LOW INSTEAD OF .SAV
;IF 2-SEG PROGRAM
3.1 Running TATTLE When expanded files have been created for each
cusp on which snooping was done, assign LPT: or spool it, and run
TATTLE. It will ask:
HISTOGRAM FILE:
The user should respond with the name of the histogram file
written by SNOOPY (defaults are DSK:MONITR.HST). TATTLE will
print the symbolic histogram listing, and return to the question.
When all desired files have been processed, type <CR> to return to
monitor mode.
PROGRAM SPECIFICATION -- SNOOPY/TATTLE Page 11
3.2 TATTLE Errors The following error messages may be printed by
TATTLE:
?CANT INIT
HISTOGRAM FILE:
The INIT UUO failed when TATTLE tried to read the
histogram file.
?CANT LOOKUP
HISTOGRAM FILE:
The LOOKUP UUO failed when TATTLE tried to read the
histogram file. Type the device, filename, extension, and
PPN of the histogram file.
?CANT INIT dev
WHERE IS EXPANDED FILE?
The INIT UUO failed when TATTLE was trying to read the
symbol table. Type the file specification for the
expanded file containing the symbol table, or <CR> to list
the histogram without symbols.
?LOOKUP FAILURE ON dev:file.ext
WHERE IS EXPANDED FILE?
The LOOKUP UUO failed when TATTLE tried to read the symbol
table. Respond as above.
?CANT FIND SYMBOL TABLE IN dev:file.ext
WHERE IS EXPANDED FILE?
The file in which TATTLE expected to find the symbol table
does not have one. Respond as above.
?INSUFFICIENT CORE TO READ dev:file.ext SYMBOL TABLE
The symbol table of the specified file is too big to fit
in the available core. The listing of the histogram
continues without symbols.
HISTOGRAM FILE FORMAT IS INCORRECT
The specified histogram file is not in the proper format.
This may be due to different versions of SNOOPY and
TATTLE, or may be caused by specifying the wrong file to
TATTLE.
HISTOGRAM FILE IS TOO BIG. ENLARGE DATA TO nnnn
The histogram file is larger than the space allowed in
TATTLE for it. Change the dimension of the array DATA in
TATTLE to at least the number given. (also change the
call to RHIST at statement 100, and the IF statement about
14 lines below it.)
[END OF SNOOPY.DOC]