Only a part of the answers (any search engine should yield plenty of infos):
1. Use proc timeseries for time series diagnostics (find trend or seasonality). You can do time series diagnostics with proc arima as well, but also use it for forcasting.
2. White Noise means, that there is no "information" (left) in a time series - values have zero mean, a non-zero variance and are serially not correlated (check wikipedia or so for full definition). I don't really understand the 2nd part of your question.
3. Plenty of stuff on the internet/any time series text book (search for ar(1), ma(1), ar(2), ma(2))
4. Maybe like this: 1. Check for unit-roots (check if time series is stationary; search "Dickey-Fuller-Test") 2. Use ACF/PACF for preliminary model 3. Adjust model if necessary (maybe use AIC/SBC as criterion - smaller is better) 4. Make sure the residuals are White Noise 