HP 3000 Manuals

HP RXForecast Users Manual for MPE Systems

Examining a Response to Prompt Case 

During the night shift on Trapper, users have perceived that performance
has been sluggish or erratic at times.  You want to determine from the
SCOPE data if you can verify this perception and, if so, see if you can
forecast what might happen in the near future.

RXForecast Options Selections 

For this example, set the options in the RXForecast Options command
dialog box as follows:


            

   *   Seasonality Level.

       All of the seasonality boxes are checked.

   *   Confidence Level.

       Set the confidence level to 90% Confidence.

   *   Interval Type.

       Choose Confidence Interval.

Forecast Selections 

   *   Metric.

       Select a metric that will measure this phenomenon.  Choose
       response time, specifically Response to Prompt.

   *   Dates.

       You are not sure when this behavior began, so use all 6 months of
       data available for Trapper.  Expect to see response time
       increasing over those 6 months if a problem is developing.

   *   Shift.

       Since the problem is more noticeable at night, select the 4 hours
       between 12:00 AM and 4:00 AM.

   *   Ignore Weekends.

       You are not interested in weekends, so choose Ignore Weekends.

   *   X-Axis.

       Choose an X-Axis length of 1 year since you are forecasting from 6
       months of data.

   *   Points Every.

       Choose Points Every Day.

   *   Trending Method.

       Select the Smoothing method in order to preview the data.  You
       should be able to see if any trend exists, and if so, if it is
       similar to any of the trend selections available within HP
       RXForecast.

Resulting Graph 

Run the model and get the following graph:


            

Analysis 

Notice that the smoothed line indicates an exponential shape to the
trend.  Close the graph.

Reforecast 

Draw the graph again, but this time use the exponential trending method.

Forecast Selections 

Return to the Forecasts command dialog box and change the following
option:

   *   Trending Method.

       Choose the Exponential trending method.

Resulting Graph 

Run the model and draw the following graph:


            

Review and Validate 

The first thing you do to validate the forecast is investigate the graph.
Notice that the data tracks well with the exponential shape.

For a more rigorous validation step, Zoom Stats on the graph to produce
the statistical report.  Notice that the Slope and Intercept parameters
are very significant, both at 100 percent, although the R-Squared value
is somewhat low at 46 percent.

Even though the correlation could be better, it is obvious that an
exponentially increasing trend exists.  The 12:00 AM to 4:00 AM shift is
not a critical one, but it could start to have repercussions on jobs that
run after 4:00 AM. Therefore, you should investigate the trend in more
detail to determine if any corrections can or should be made.

Reforecast 

You can take one or two approaches to determine if most of the increase
in response time can be attributed to a specific application.  If you
could determine that a specific application is responsible for most of
the increase, you would have better information for solving the problem.

You could repeat the forecasting process, but in greater detail.  Instead
of plotting Global Response to Prompt, you could plot Application
Response/Prompt for every application.  You would plot all configured
applications with a Smoothing trend for the same time frame and shift as
before.  Eventually, you would find that the OTHER application displays
an exponentially increasing trend:


            

Another approach to discovering that the OTHER application is responsible
for most of the increase in response time would be to use HP LaserRX/MPE.
You could use the Application Command under the Draw menu to display the
Application Transaction Response graph.  The spike on the graph indicates
that the OTHER application is the busiest application.


            

Resulting Graph 

Whether you use HP RXForecast or HP LaserRX/MPE to pinpoint the OTHER
application, the next step is to plot a forecast using the Exponential
trending method.  The following graph is produced:


            

Review and Validate 

Generate the Stats report.  This time the R-Squared is a bit higher at 49
percent and, as before, the Slope and Intercept parameters are at 100
percent significance.

At this point, you cannot get any more specific details on why the trend
is heading upward exponentially.

You do not have any process data in the logfile, so it is impossible to
determine exactly which programs are experiencing these growing response
times.  You are not certain what corrective action is required, but you
are certain that a trend is developing, according to the HP RXForecast
analysis.

You would proceed by finding out what applications make up the OTHER
application and reconfigure the PARM file.  You would also turn on
logging of process records for that shift to obtain more detail on the
cause of the increasing response times.  This case underscores the
importance of effective data management.

MPE/iX 5.0 Documentation