HP 3000 Manuals

HP LaserRX/MPE: A Journey of Discovery

Taking the Train to Response Times 

What about response time?  Many users measure system performance by this
statistic alone.  They consider any system that does not respond quickly
enough to be slow, regardless of what else the system might be doing.

Go back and check the terminal response times on Trapper.  To do this:

   1.  Close all active windows, either individually or by playing back
       your macro.

   2.  Return to the Draw Graphs dialog box, and select the following:

          a.  Graph=Global Transaction Response.

              Remember to deselect Global Disc Detail.

          b.  X-Axis=Year.

          c.  Points Every...=Day.

          d.  Shift=All Day.

          e.  Starting Day=1 March 1988.

   3.  Click  OK.


          Global Transaction Response Graph 

What does this graph show you?  The terminal response times increased
markedly around 1 June 1988.  The average time to achieve the first
response jumped from less than 1/2 second to more than 3 seconds.  The
response to prompt jumped from about 2-1/2 seconds to 10 seconds.  The
average transaction rate, however, remained very low at a steady 200
transactions per hour.  (Busy systems can run as high as 6000
transactions per hour without a problem.)

Response times are affected by the entire system, not only by a specific
piece of hardware like the CPU or disc.  You need a slightly different
approach to study response times.  Here, you know that response times
changed drastically, but which transactions changed?  Did a single
application's transactions change, or did all transactions slow down?  To
check this, do the following:

   1.  From the Draw Graphs dialog box, select the following:

          a.  Graph=Application Transaction Response.

              Remember to deselect Global Transaction Response.

          b.  X-Axis=Week.

          c.  Points Every...=Hour.

          d.  Shift=All Day.

          e.  Starting Day=1 March 1988.

   2.  Click  OK.

Application graphs are drawn based on the application definitions set up
by the user.  They are similar to Disc Detail graphs because both types
of graphs show the top five user-defined applications sorted in
descending order based on the main item in the graph.  In this graph,
applications are sorted by transaction rate; in other application graphs,
they might be sorted by CPU. One curve is drawn for each application to
show how its transaction rate varied over the selected time period.

Use the vertical scroll bar to select an application for detailed study.
You can do this by moving the scroll box until the desired application is
noted in the dialog box, or you can click the arrows in the vertical
scroll bar to move between applications.  Notice that you can select any 
defined application, not only one of the top five shown.

This graph confirms that the transaction rate is quite low, and it tells
you that most of the transactions are being processed within the
catch-all application called Other.


          Application Transaction Response Graph for Other 

Use the vertical scroll bar to plot more detail on the Other application.
(Click the vertical scroll box, and with the mouse button depressed, drag
the box until the dialog box in the center of the screen says Other.
Then release the button.)

The original graph is replaced with the Application Transaction Response
graph.  It looks similar to the Global Transaction Response graph, but it
only shows transactions that belong to the selected application (in this
case, Other).  Be sure that the response times on this graph match the
first part of the Global Transaction Response graph when response times
were good.

Next scroll the Application Transaction Response graph horizontally to a
date that falls after the change in response times--such as 15 August
1988.  Check the response times for this application; they should
correspond to those on the second half of the Global Transaction Response
graph (when things looked bad).

Extra Credit Exercise 

Scroll to other applications' response times before and after the change
in overall response times, and see if their response times varied in the
same way as the Other application's did.

You know that the application having problems is called Other.  Now what
can you do?  You need more details.  You have to examine individual
processes within the Other application, but you cannot do that now.  When
the TRAPPER.PRF file was created (extracted), process data was excluded
to save disc space.

MPE/iX 5.0 Documentation