14. WORKLOAD FORECASTING › 14.2 Usage Guidelines › 14.2.2 Choosing Model Parameters

14.2.2 Choosing Model Parameters

Choosing the model parameters for the Workload Forecasting
application using Neugents technology is reasonably simple.
Aside from the Primary CAP file and Analysis element
specifications, the CA MICS Capacity Planner software
provides a set of default values that will permit a valid
model to be obtained with minimal effort.

Building a Neugents technology model can be and often is an
iterative process.  The model is always trained first and a
forecast produced according to the Forecast Length
specification.  Descriptive statistics of the input data, and
model performance statistics from both the training and
forecasting phases are produced.  Additionally, a graph
depicting the results of the training and forecasting phases
is generated.

Using the default Forecasting Method of AUTOMATIC will cause
a model to be based on the historical growth of the values of
the Training elements, and should provide a good starting
point for capacity planning. Once better information
concerning growth rates is available, then the Forecasting
Method can be set to MANUAL and specific adjustment factors
used for each selected workload.

When specifying a Forecast Length value, use something
reasonable that relates to the business requirements.
Neugents technology is quite adept at forecasting, but any
forecast begins to lose validity when the period is
exaggerated. Depending on the quantity and granularity of the
input data, a forecast period of 90-180 days is probably a
good place to start. Save your forecasts and compare them to
the actual data when it becomes available. As with any
forecasting technique, the validity of the forecast is a
function of length of time covered by the input history data.
For example, we would never suggest using six months of
monthly summarized data and then attempting to forecast 18
months into the future. The Neugents technology model would
produce a forecast, but it would be of no practical value.