Growth analysis I am not a statistician but a Java/R programmer. So even the topic could be wrong. Please bear with me.
I am collecting details from production servers. These details could be the number of active sessions, CPU utilization etc. I draw graphs. So for example, I have a R graph showing Time(x), Web Server hits(left y-axis) and MBytes transferred(right y-axis). Another one is the number of active sessions over time.
I need to understand the statistical growth pattern of this data before and after an event. These are all distributions.


*

*How do I go about measuring the growth ? I use R.

*How do I understand what causes that growth ? I think this is about regression.


The more complex question from my perspective.


*

*What is the statistical process to predict future growth in these cases ? I already read Capacity Planning books. This is not about Capacity planning which probably is the next step.


I have come across topics like 'Growth Analysis curves and Visualization' but couldn't access any material even after searching.
Mohan
 A: The easiest approach to growth modeling is simply handling time as a fixed effect in a regression model, such that time (or some derived measure(s)) is your X and the outcomes are your Y. From there, several flexible tools can be used to show pre/post differences. 
You may simply recode time as 0 for any time event before the intervention and 1 for any time after. You may also code time continuously to inspect secular trends of increase or decrease. Blending that you can have both such terms as well as their interaction for a rudimentary breakpoint analysis. More sophisticated methods include using splines. 
Once you are satisfied with how you have measured growth, to find if certain variables affect growth (such as work from home agreements, e.g.), you can add these as time-varying or between cluster variables in your model; and, more importantly, you add the interaction between these variables and the growth terms. If the interactions are statistically significant, you would conclude that they affected growth as you say.
A somewhat different approach is taken to "predicting" growth--which (if I understand your problem) time-series modelers call forecasting. (The distinction is that you are not predicting growth per se, but using measures of growth to predict outcomes later in the future). The problem with the previous methods is that fixed effects tend to "explode" with time, leading to silly forecasts such as Olympic lap times of 0 seconds by year 2030. ARMA or ARIMA models are very standard tools for forecast models which are simple, flexible, and easily computed. These work by including previous outcome measure(s) as covariate(s) in a predictive model. Once new results are realized, forecasts may be updated to account for stochastic variability.
A text which I have read on the topic that I found useful is Longitudinal Data Analysis by Diggle Heagerty Liang and Zeger, which is not an open resource admittedly. No satisfactory discussions have I encountered on the subject.
