HP 3000 Manuals HP 3000 Manuals
Producing the Forecast
This step in the forecasting process requires choosing the following:
* Certain HP RXForecast options.
* Seasonality Level.
* Interval Type.
* Confidence Level.
* The forecast length (X-Axis).
* The Trending Method.
![[]](../../pic3k/M011760/FFN273c50.gif)
Forecast Length
You choose the length of the forecast through the X-Axis option. In HP
RXForecast, you can choose an X-Axis of Quarter, 6 Months, Year, 2 Years,
or 3 Years.
Choose an X-Axis that is longer than the time span you choose with the
Dates option (Start and Stop dates). The Start and Stop dates determine
the length of the actual data from which you generate a forecast. If the
Dates time span is as long as the X-Axis, no room will remain on the
graph's X-Axis for displaying a forecast.
Think of the X-Axis as consisting of two components: raw data and
forecast. It should be long enough to accommodate both. The part of the
X-Axis that is not filled with raw data will show a forecast. For
example, if you have 6 months of data, use an X-Axis of 1 year, since
this will yield a forecast of 6 months.
Tip Forecasts that span a longer range than the amount of data you have
are not reliable. Thus a good practice to follow is to limit the
length of the X-Axis to double the length of the data. For
instance, if you have 6 months of data, limit the X-Axis to 1 year.
Trending Method
Generally speaking, you would choose Smoothing as your first trending
method. Smoothing allows you to preview the data. On a smoothed graph
you can more readily detect the presence of a trend. Data graphed with
Smoothing may behave approximately as one of the other trends (Linear,
Exponential, S-Curve).
NOTE If you select the Smoothing method, the part of the X-Axis that is
not dedicated to plotting raw data will be blank. This method does
not generate a forecast.
If you see a Linear, Exponential, or S-Curve trend on a smoothed graph,
choose the appropriate trending method on the HP LaserRX/MPE Forecasts
command dialog box to produce the forecast.
The kinds of trends that you would try to identify are the following:
Linear:
![[]](../../pic3k/M011760/FFN283c50.gif)
The diagram above illustrates the kind of pattern in which the Linear
method is appropriate. Not all of the points have to lie on a straight
line. Linear trending is appropriate if a straight line can reasonably
represent the set of points. The Linear method is the most commonly used
trending method.
Exponential:
![[]](../../pic3k/M011760/FFN293c50.gif)
The figure above shows the general patterns of data where exponential
trending may apply. Notice that the trend moves upward (or downward)
steeply along a curve. This distinguishes the Exponential method from a
Linear or straight-line method.
S-Curve:
![[]](../../pic3k/M011760/FFN303c50.gif)
The S-Curve, as the name suggests, is shaped like the letter S. Moving
from left to right, the S-Curve starts out as a slightly increasing curve
that gradually increases in slope until it rises sharply like an
increasing exponential curve. Finally, it levels off and approaches a
horizontal line. A reversed letter S would indicate a decreasing curve.
Tip The CPU utilization curve may look like an S-Curve with the middle,
sharply increasing section rising to about 90 percent before
leveling off and slowly approaching 100 percent utilization.
The S-Curve trending method would be inappropriate for forecasting
response time, since response time does not have an upper limit.
If the time period does not display steady change according to a Linear,
Exponential, or S-Curve trend, then the only viable trending method
remaining is the Business Units method. If you can correlate changes in
system performance to changes in a business unit, you can generate a
valid forecast.
See Chapter 4 for details on creating and using a business unit file.
Other Options
Before you produce a forecast, you might want to change the following HP
RXForecast options:
* Seasonality Level.
* Interval Type.
* Confidence Level.
Seasonality Level.
When you start HP RXForecast for the first time, the seasonality levels
on the RXForecast Options command dialog box are all enabled. You can
change these default settings for one session or every session of HP
RXForecast just as you change configuration options in HP LaserRX/MPE.
When any seasonality level is checked along with Auto Seasonality, HP
RXForecast calculates the forecast and checks to see if the calculated
seasonality is significant. If it is significant, it is included in the
forecast; if it is not significant, it is excluded. Thus, for the most
part it makes sense to check all seasonality levels because as long as
you have Auto Seasonality checked as well, HP RXForecast will decide
whether or not to include it.
Tip Seasonality may be forced into a forecast by checking any or all
seasonality levels, but leaving Auto Seasonality unchecked. Be
careful when interpreting the apparent seasonality of a forecast if
you have selected these options.
Interval Type.
You should decide if you want to change the interval settings.
While in the RXForecast Options command dialog box, change the Interval
Type option if appropriate. Choose a Confidence Interval if you want to
see where 90 percent of all true mean values lie. Choose a Prediction
Interval if you want to see where 90 percent of all values fall. Choose
None to display no interval type at all.
Tip If you choose Confidence Interval and have trouble reading the
resulting forecast, try reforecasting with Prediction Interval. It
is wider and easier to read.
Confidence Level.
You should also decide if you want to change the Confidence Level option.
The default level is 90% Confidence. If you prefer, you can choose 70%
Confidence or 80% Confidence.
After you have made your choices on the RXForecast Options and Forecasts
commands dialog boxes, you are ready to have HP RXForecast calculate a
forecast and display it on the screen. You do this simply by clicking on
OK on the Forecasts command dialog box.