Trailing-Edge
-
PDP-10 Archives
-
decuslib10-06
-
43,50437/boss.mem
There are no other files named boss.mem in the archive.
Care and Feeding of
BOSS
A Method of Reporting System Statistics
Date: 15-Oct-76
File: BOSS.RNO
Author: M. Barnes
BOSS Page 2
Abstract
Abstract
The BOSS system, consisting of BOSS, SYSREP, WEEKLY and
PLTSYS, provides an easy method of reporting system
statistics (such as the number of UUO's per second) which
are unattainable from the system fact files.
System Overview
BOSS (the program) runs by itself (usually under batch)
and gathers system information at half hour polling
intervals. It appends this information to a binary file.
SYSREP uses the file produced by BOSS and creates three
reports: SYSREP.DAT, DSKREP.DAT, and SCHREP.DAT. It
optionally makes PLTDAT.DAT for use by PLTSYS, and always
makes an *.WEK (* is the same file name as on the .BOS file
being used) file for use by WEEKLY.
WEEKLY uses a series of *.WEK files to create
WEKREP.DAT, a daily summary report, containing one line per
day, for a period of days specified by the user.
PLTSYS uses PLTDAT.DAT or an *.WEK file to create a
plot file for plotting on a Calcomp 11 inch drum plotter.
Notes
The entire BOSS system is written in Fortran-10 and
Macro. BOSS also requires the services of IPCF routines
(also written in Fortran and Macro) to insure that only one
BOSS is running at a time.
Note that the BOSS system was written for the KL-10 at
UTHSCD(MCRC) and gathers some information especially for the
KL. Also note that if your system has more or less than
five disk drives (see Known Bugs), certain parameters in
SYSREP will need to be changed (such as array sizes).
BOSS Page 3
More on BOSS
How to run BOSS
To run BOSS all one needs to do is type RU BOSS. BOSS
runs by itself with PEEK privileges. BOSS creates the file
*.BOS (where * is the left half of the universal date/time)
and appends to this file every half-hour. BOSS relies on
IPCF to make sure that only one BOSS is running at any one
time. At twelve midnight BOSS changes file names (to the
new date). To run BOSS under batch see 'BOSS and Batch'.
BOSS Page 4
Using SYSREP
Using SYSREP
SYSREP prepares the files SYSREP.DAT, SCHREP.DAT, and
DSKREP.DAT for queueing to the line printer, *.WEK for
WEEKLY, and optionally PLTDAT.DAT for PLTSYS. SYSREP reads
as input the files created by BOSS. When SYSREP asks for
the date, type in either the file name (not the ext) created
by BOSS or 'YDAY' or 'TDAY'. YDAY stands for yesterday and
TDAY stands for today.
If 'YES' is typed in response to the query about making
plot files, PLTDAT.DAT is created. Otherwise it will not
be.
BOSS Page 5
Using WEEKLY
Using WEEKLY
WEEKLY prepares the report WEKREP.DAT using .WEK files
created by SYSREP. When run WEEKLY will ask:
Last File Name:
Type one of the following:
<6 digit octal number (*)> - to start with *.WEK
WEEK - to do seven days ending with yesterday
2WEEK - to do fourteen days ending with
yesterday
MONTH - to do thirty days ending with
yesterday
If your response to the above was a six digit octal number,
WEEKLY will ask for the number of days to process.
BOSS Page 6
Using PLTSYS
Using PLTSYS
PLTSYS will query for the file name to plot information
from ('FILE(No ext): '). Input either a <cr> or 'PLTDAT'
to read PLTDAT.DAT (created from the last SYSREP in which
plotting was specified), or a six digit octal number to read
*.WEK (* is your octal number), created from *.BOS when
SYSREP was run. PLTSYS asks you what plots you want; if
you want a particular plot, type 'YES'. In addition, if you
type DEFAULT to the first plot query made by PLTSYS, PLTSYS
will read the file DEFAUL.BOS to determine which data to
plot. The default file contains one line for each plot.
The plot codes for DEFAUL.BOS are as follows:
LOGINS number of logins vs time
UUO/SEC number of UUO/SEC vs time
CTSW/SEC number of context switches per second
COMMANDS number of commands vs time
IDLE cpu idle time(%) vs time
LOST cpu lost time(%) vs time
USER user time(%) vs time
OVHD monitor overhead time(%) vs time
LINES number of active lines vs time
SWAP number of swap I/O's vs time
JOBSIZE average job size (pages for KI,KL)
SWAPPER (%) swapper null time
REQUEUE number of requeues per second
JBORDER number of jobs run out of order
After PLTSYS has created a plot file (*.PLT: * is a
spool name), it will exit if the input file was PLTDAT, or
if not, start over and ask for a new file name to read in.
To exit from this type 'STOP' (NB: if your system does not
have auto-spool and you make multiple plot files, they may
overwrite each other because of FOROTS lack of originality
in spool names).
BOSS Page 7
Installation Procedures
Installation Procedures
The only major change in the code for different
installations is the number of disks. Here at MCRC we have
five. Your installation may differ, so make the following
changes (see Known Bugs):
In SYSREP.PRM change NUMDSK = 5 to the number of
disks at your installation.
In CHKNAM.FOR change PPN = ? to the PPN that BOSS
is to be run under.
To change to scale setting for plots in PLTSYS,
change the calls to SETSCL. The format is 'CALL
SETSCL( <max number of units (%,swap I/Os,etc)> ,
<number of inches (height) to plot in> ). Each
individual plot has its own subroutine and call to
SETSCL.
BOSS Page 8
Compiling and Loading
Compiling and Loading
Please follow the below descriptions for compiling and
loading the BOSS system:
.COM/COM/CREF @SYSREP.COM
.COM/COM/CREF @BOSS.COM
.COM/COM/CREF @PLTSYS.COM
.COM/COM/CREF @WEEKLY.COM
.R LINK
*@SYSREP.LOD
*/SA/G
.R LINK
*@BOSS.LOD
*/SA/G
.R LINK
*@PLTSYS.LOD
*/SA/G
.R LINK
*@WEEKLY.LOD
*/SA/G
.CREF
.PRINT *.LPT
BOSS Page 9
The BOSS system and Batch
The BOSS system and Batch
The recommended method of running BOSS (the program),
is to use batch. This way if the system crashes
(unfortunately a much too frequent occurrence), batch can
restart BOSS automatically. The control file to run BOSS is
BOSS.CTL, and to get it started, use a Submit command such
as:
SUBMIT/UNIQUE:0/RESTART:1/TIME BOSS
Reports may also be generated by using batch. The
control file to do this is BOSOUT.CTL. BOSOUT will run
SYSREP, making the reports for the previous day, then
WEEKLY, and finally PLTSYS. It queues all reports
(???REP.DAT) to the line printer and the plot file made by
PLTSYS to the plotter. It also submits itself for the next
day. BOSOUT should run at about 00:05 AM (just after
midnight to get the complete *.BOS file), and should be
submitted initially with a command string similar to:
SUBMIT/UNIQUE:0/RESTART:1/AFTER:00:05 BOSOUT
BOSS Page 10
Known Bugs and Deficiencies
Known Bugs and Deficiencies
SYSREP
Disk statistics work fine for N (N = any number) number
of disks on controller 0, or N disks on controller 0 and one
disk on controller 1. If you have two or more controllers
(if two controllers, then more than one disk on controller
1), then disk statistics will not work properly. You will
have to fix HND601 to suit your system.
Certain data (Cache, MBox, EBox) is gathered for the
KL-10 (single processor 1080 system). This should not
effect anything but SCHREP.DAT, and only to the extent that
zeroes instead of valid numbers will appear under KL-10 only
headings. This also applies to older monitors that do not
collect scheduler data (Average job size, Jobs run out of
order, Requeues/sec, % Swapper null).
BOSS
Please attempt to have DAEMON running before BOSS.
BOSS will run either way, but DAEMON first is suggested.
Also try to have SYSINF (if it is being used) running before
BOSS.
BOSS uses IPCF in order to determine whether or not
another copy is already running (NB: Only have one BOSS.SAV
running at one time (unless they write into different
files)). If your system does not use IPCF, modify INIT.FOR
by putting a RETURN statement after LEN601 = 18. But beware
of having two BOSSes running at one time.
BOSS Page 11
SYSREP Title Descriptions
SYSREP Title Descriptions
From - Start time of interval
To - Stop time of interval
Resp Time - System response time (I)
Login - Number of users logged in (P)
Hrdw Errs - System-wide hardware error count (I)
Cmnds - Number of commands processed (I)
% Idle - Percent CPU Idle time (I)
% Lost - Percent CPU Lost time (I)
% Ovrh - Percent CPU Overhead time (I)
% User - Percent CPU User time (I)
UUOs/sec - Average number of UUOs per second (I)
CtSw/sec - Average number of times processor changed jobs/sec (I)
WdCl/sec - Average number of words cleared per second (I)
Active Lines - Number of active lines (P)
Swap I/Os - Number of Swap I/Os/100 (I)
Rcv Scanr - Number of RCV interrupts/1000 (I)
Xmt Scanr - Number of XMT interrupts/1000 (I)
CorMax - Maximum number of pages (or K) allowed (P)
Note: I - Taken over interval (usually 30 minutes unless startup).
P - Taken at polling time (To time).
BOSS Page 12
DSKREP TITLE DESCRIPTIONS
DSKREP Title Descriptions
Drive - Physical drive name
Pack - Pack name and ID
Mount Count - Number of users having pack mounted (P)
% Used - Percent of disk space used (P)
Montr Seek - Number of monitor seeks (I)
User Seek - Number of user seeks (I)
User I/Os - Number of user I/Os (I)
Montr I/Os - Number of monitor I/Os (I)
Total Seek - Monitor seek + User seek (I)
Total I/Os - Monitor I/Os + User I/Os (I)
I/Os Seek - Total I/Os / Total seek (I)
Hard errs - Hardware disk errors detected (I)
Soft errs - Software disk errors detected (I)
Note: I - Taken over interval (usually 30 minutes unless startup).
P - Taken at polling time (To time).
BOSS Page 13
SCHREP Title Descriptions
SCHREP Title Descriptions
From - Start time of interval
To - Stop time of interval
Average Jb Size - Average job size in pages (or K) (P)
Jbs out of order- Number of jobs run out of order to allow
them to give up resource required to run
a job (I)
Requeues/sec - Average number of scheduler requeues per second (I)
% Null - Swapper percent null (I)
EBox - Jiffies of KL10 EBox time (I) [KL]
MBox - Jiffies of KL10 MBox time (I) [KL]
Cache - Average number of cache requests per second/100
(I) [KL]
Note: I - Taken over interval (usually 30 minutes unless startup).
P - Taken at polling time (To time).
BOSS Page 14
WEKREP Title Descriptions
WEKREP Title Descriptions
Logins - Number of users logged in (+)
% Lost - Percent CPU Lost time (=)
% Idle - Percent CPU Idle time (+)
% Ovrhd - Percent CPU Overhead time (+)
% User - Percent CPU User time (+)
Max Swap IO - Maximum number of Swap I/Os/100 for day
Ave Resp - Average response time for day
Max Resp - Maximum response time for day
Systm Crash - Total number of reboots for day
HrdWr Errs - Total number of Hardware errors for day
Ave JbSz - Average job size for day
Ave CtSw - Average number of times processor changed
jobs/sec for day
Ave UUO - Average number of UUOs/sec for day
Ave WdCl - Average number of words cleared per
second for day
Note: = - Contains Ave,Max for day
+ - Contains Min,Ave,Max for day