Selecting Data [ HP RXForecast Users Manual for MPE Systems ] MPE/iX 5.0 Documentation
HP RXForecast Users Manual for MPE Systems
Selecting Data
The next step in the forecasting process is to select the proper subset
of data for a specific forecast. This requires choosing the following:
* The HP LaserRX logfile.
* The metric to forecast.
* The start and stop dates.
* The shift.
* Whether or not to ignore weekends.
* The summary level.
Available Logfiles
You should decide which logfile has the correct data for the metric you
want to forecast. The accuracy of a forecast is related to how
representative the data is for the variable you want to forecast.
NOTE When you choose a logfile in HP RXForecast, the same logfile will
be chosen on the HP LaserRX/MPE dialog boxes.
You may have multiple logfiles for the same computer system. For
instance, you may have one logfile that contains only Global data
summarized to the hour, another logfile that includes application data,
other logfiles with shifts other than all day, and some that ignore
weekends.
NOTE If you pick a logfile that does not contain application data, the
Application Names list box on the NEXT screen is empty.
Effectively managing your logfiles can make forecasting easier. Some
helpful hints about data management are included in "Managing Data
Effectively" .
The Metric to Forecast
You should have an objective in mind when you forecast your system's
performance. There will be times when projecting such values as CPU
Utilization, Response Time, and I/O Rate will be key to maintaining an
acceptable level of performance.
The performance of particular applications may also be important to
maintain. The selection of metrics to forecast should reflect the
specific needs of the environment. Simply selecting all metrics, with
all applications, using all possible subsets of the data and shifts is
neither productive nor reasonable.
NOTE If you pick a logfile that does not contain application data, the
metrics list box on the Forecasts command dialog box does not list
application metrics.
Start and Stop Dates
You should choose the range of data you want included in your forecast.
This range is known as the time frame and is defined by the Start and
Stop dates.
The time frame you select is important because the data you choose is the
only information that HP RXForecast has available for making a forecast.
A forecast is only as good as the data upon which it is based. In
general, the accuracy of a forecast increases with a longer time frame.
NOTE When you choose the Start date in HP RXForecast, HP LaserRX/MPE
will display the same dates in its dialog boxes.
You should avoid choosing a time frame that spans or includes any of the
following:
* Hardware changes.
* Configuration changes.
* Workload redefinitions.
Choose time periods of stable activity.
Tip Your raw logfiles may contain data from periods of stable activity
that are broken by an unusual activity, such as a holiday. Rather
than ignore such a period, use the EXTRACT program to create a
logfile that eliminates the unusual activity. See "Managing Data
Effectively" .
Shift
The shift is the part of the day in which you are interested. Thus the
shift is always a time period of less than 24 hours and is often 8 hours
or less.
You use the Shift option on the Forecasts command dialog box to define
the shift. The shift can cross over from one calendar day into the next
(for instance, 9 PM to 7 AM).
NOTE When you choose a shift in HP RXForecast, HP LaserRX/MPE will
display the same shift in its dialog boxes.
The shift you select should reflect your purpose in making the forecast.
If you are interested only in the peak period of the day, then you should
choose the peak period shift. If you are interested only in performance
when key applications run, then you should choose that shift. If you are
interested only in average performance, you should use the entire day as
your time window (realizing that there is a danger that doing so could
produce results that indicate very poor performance during the peak
period of the day).
Ignore Weekends
Weekends may or may not be an interesting part of your work week. Some
systems or metrics have peak demands during the week, with very little on
the weekend, while others peak on the weekends. This is a decision that
only you can make. By selecting Ignore Weekends, you can exclude weekend
data from the graph.
Tip If your business runs on the standard 5-day work week, you might
want to get in the habit of ignoring weekends. Otherwise, you will
see day-of-week seasonality in your forecasts, which needlessly
clutters them.
Summary Level
The summary level--defined through the Points Every option--affects the
variability that you see on a forecast. For instance, summarizing to the
week eliminates the variability of a daily summary.
What summarization level you choose depends very much on how much data is
in the chosen logfile. You would not choose Points Every Week if you
have only 1 month of data.
Choose a Points Every option as you would in HP LaserRX/MPE.
NOTE When you change a Points Every Day, Week, or Month option on HP
RXForecast, it automatically changes to Points Every Day in the HP
LaserRX/MPE dialog boxes since Day is the highest summarization
level in HP LaserRX/MPE.
MPE/iX 5.0 Documentation