.OPTION INCLUDE FULLTOOLS
.ce
AFDSC MANAGEMENT REPORTING SYSTEM
.sp
.PP1
This paper describes the Air Force Data Services Center (AFDSC)
Management Reporting System.
The system was born out of a management problem:
.PP2
Over three thousand AFDSC customers (including 800+ UNIX
.FS
UNIX is a registered trademark of Bell Laboratories.
.FE
users)
were getting good and bad impressions of ADP on a daily basis.
Truth, rumors, priorities, and "politics" prevented reliable
feedback about user satisfaction to management.
.PP2
There was no easy means to tell if system service level was high or low
(i.e., the customers were right or wrong in their complaints).
There was no easy means to compare today's performance with that of past days.
There was no 
.ul
common
data collection among different OS's to allow performance comparison.
.PP2
There was no means to benchmark system "enhancements" in a real user environment.
Most operating system (OS) diagnostics were 
.ul
too
informative and specific.
For UNIX, there were no OS diagnostics to even start with.
.PP1
The AFDSC-wide solution encompassed many features:
.PP2
A daily report was standardized for 
.ul
all
AFDSC systems (IBM, GCOS, MULTICS, UNIX).
.PP2
Only four areas were presented:
Performance (response time);
Availability (up and down times, number of crashes);
Utilization (cpu time, login hours);
and Workload (number of users and commands).
.PP2
Yesterday's data was compared with that of the past 30 days.
.PP2
Summarizations, averages, max and min values were displayed for
the 30 days prior, to aid in comparisons.
.PP1
The UNIX Implementation involved:
.PP2
.ul
No
kernel changes (/dev/kmem is read from without).
.PP2
A "probe" program to access the running kernel and take snapshots of system
status and tables.
.PP2
A "mgt.sum" program to merge login, shell, and "probe" accounting files;
compute response time using "probe" accounting and synthesis technique;
compute "user support factor" (number of users this system can support).
.PP2
A "mgt.rpt" user command to print a daily management report or a kernel
"high-water" report (showing utilization of system resources such as tables).
.PP1
The benefits we derived were numerous:
.PP2
Automatic tracking of crashes that are usually difficult to trap by automated means.
.PP2
Consistent, reproducable analysis techniques.
.PP2
Documentation of "enhancements" during real use (versus "gut feeling").
.PP2
Reliable sizing of system tables.
.PP2
Measurement of system size versus user community size.
.PP2
Reporting system can be disabled without internal changes.
.sp3
.nf
Walter D. Lazear and Charles W. Muir

Air Force Data Services Center
The Pentagon, Room 1D988
Washington, DC 20330

Commercial (202) 695-6161
Autovon 225-6161
.fi
