Probe/iX: An Online
Advisor for 3000 Performance
Analysis tool trades
overhead hit for plain-English reports on performance
Review by Shawn Gordon
Probe/iX is system
performance monitoring software for the HP 3000. Not a glamorous or
exciting category rather like security but a critical one for
proper system performance. Through a standard, terminal-based interface,
Probe makes use of various line drawing and enhancement escape sequences to
present its data in a comprehensive fashion. Through function keys and
keyboard short-cuts, you are able to quickly navigate through the
How does it work?
Probe/iX uses the AIF
Measurement Interface (AIF:MI) to collect its system information (this is
how everyone does it, other than the Shot Nugget program from Allegro). The
program has a configurable scan rate which determines how often it collects
the data to display. There is a conflict here: the longer you let it run,
the more accurate trend information you get but tests have shown
that enabling the AIF:MI can consume between 10 and 40 percent of the CPU
above and beyond the amount that it reports on itself.
you usually want to run a tool when the system is slow, to see what the
problem is, you will only make the problem worse by looking. Its
probably the only way to find out what is going on. As I said, this is true
of any tool that uses the AIF:MI; there is an expense associated with using
Hopefully that talk
about the overhead didnt scare you off, because these are great
tools. Probe/iX is broken down into the sections and sub-sections listed in
|Overview (see Figure 1)
||Disc Queue Lengths
|Overview (see Figure 2)
|Detail (see Figure 3)
The Workload section of
Probe/iX allows you to lump different users and processes into logical
groups so that you can track, display and record performance information
for that entity. So if you wanted to have data entry grouped together, you
would just assign the user IDs or the program name to a workgroup. This is
a very handy facility for tracking the impact that your different groups
have. However, if you are using a 4GL, its not as useful, since
everything is charged to the single 4GL program that is interpreting your
code but you can still make use of the user tracking.
information is very nice. It will instantly (or almost) display a list of
all databases that are currently open. You can then select a database to
see all the users that have it open, and with what process and
There are function key
options within most of the screens that give you access to other
information, such as the Online Analyst seen in Figure 4. This gives a very
English overview of what is going on with your system, and
suggestions on how to improve it. I love this section.
Your CPU is 16.0% busy, which is HIGH for your system by
-Processes are Paused for I/O 576% longer than normal.
-System Processes are using 198% more CPU than normal.
-The Dispatcher is using 366% more CPU than normal.
->The CPU Hog is PIN #159 who is using 8% of the
(#S45 SHAWN,MGR.SMGA running CI.PUB.SYS)
* * *
Your Memory Manager Activity is HIGH for your system.
The Memory Manager is using 879% more CPU than normal.
-The Mem Mgr is using 100% more CPU than normal fetching
-The Mem Mgr is using 779% more CPU than normal on swap
-The Mem Mgr is doing 370% more Prefetches than normal.
-The Mem Mgr is doing 366% more Posts than normal.
-The Mem Mgr is doing 38% more Recover Overlay Candidates
-The Mem Mgr is doing 276% more Swap Operations than
*** I noted higher-than-usual Recover-Overlay-Candidates,
Swaps or IMIs...
*** One or more of these over a sustained period is usually
*** of a SERIOUS Memory Shortage problem!
->The Memory Hog is PIN #2 who is using 41.65
megabytes of real memory.
(# 0 ,MANAGER.SYS running LOAD.PUB.SYS)
* * *
Average Disc I/O rate is 4.0/sec which is HIGH for your
system by 96%.
Average Disc Queue Length is 0.3 which is LOW for your
I noted an Average Response Time of 0.87 seconds, compared
to your normal
Transactions are 434% higher than normal.
* * *
Switches to NM are 172.3/second, which is HIGH for your
system by 44%.
Switches to CM are 39.3/second, which is HIGH for your
system by 354%.
Current Process Stop Rate is 90.6/second, which is HIGH for
your system by
Current Process Launch Rate is 44.9/second, which is HIGH
for your system
Your Launches-to-Stops ratio is 0.5-to-1.
You can also change the
priority of executing processes by either putting them into another queue
or giving them a linear priority. If you take a look at the different
examples, you will see just a small part of the wealth of information
available to you.
Installation is your
standard restore, stream, put tape back in. Everything works without a
hitch. The manual is a little over 140 pages and is very well written.
There are some terrific sections at the back that go into serious detail
about performance and how to tune a system. This includes such things as
the most efficient types of counters in COBOL (or any language), how to
declare and use integers, database design considerations, disc space usage
and file spreading for performance, load management, memory management, CPU
management, etc. This is really an excellent manual for both using Probe/iX
and learning more about system management in general.
I spent a little time
loading up my poor 922 and testing Probe, then decided to give it a run at
a client site that had many more people on it. I was able to use the
displays to help them justify a memory upgrade using the Online Analyst.
This provided them, essentially, with an unbiased opinion of the system,
which is just the push that some people require.
Another nice feature of
Probe/iX is that if you run it in a process-handling environment such as
MPEX or QEdit, when you exit it will suspend itself instead of terminating.
This is nice, because the program will start right up again without the
wait time associated with its normal startup and I believe it
disables all the high overhead stuff when it suspends, such as the
Basically, I just had
fun running the product and shuffling processes around in the queues to
increase system performance during some peak usage. We were able to
identify a couple of real hog programs that we dedicated some resources to
fixing as well.
The obvious comparisons
are to Glance from HP and SOS from Lund Performance. Probe has been around
longer than either of these products, although it seems to have gone into
hibernation for the last five or six years. Probe is far superior to
Glance, and is probably similar to SOS. Both Probe and SOS offer features
that the other doesnt, but they both include all the basics. I love
the Online Analyst feature in Probe, which is even more English-like than
the suggestions provided in SOS.
Using Probe again was
like catching up with an old friend. I had first used Probe back on MPE/XL
version 1.2, and it was my first experience with a third party tool
designed to look at a system. Prior to that there were all the CSL versions
of Overlord and its many offspring. I like Probe a lot, a whole lot. The
product provides information that none of the competitors do, such as open
databases and whos got them open. It seems to be priced less than the
competition as well. The interface is easy and the screens provide a wealth
of information (you just have to understand what its telling you).
The Online Advisor helps tremendously for people who either have trouble
understanding the displays, or have trouble explaining it to someone else.
You know, those discussions you have while buying hardware upgrades.
If youre one of
the few that hasnt actually purchased an HP 3000 system performance
monitoring tool, then add Probe/iX to the list of products to compare.
Shawn Gordon, whose S.M. Gordon & Associates firm
supplies HP 3000 utilities, has worked with 3000s since 1983.