Ask HN: Would a Markov Model be appropriate in this situation?

Bandwidth usage patterns often occur over much longer periods of time. As a result, while a markov model could be used, you'd need a large order, and you'd need to give it a lot of data (like, ten to fifteen years because bandwidth patterns occur on a week or year scale in many cases). Obviously, this is a tall order -- many organizations haven't even been online for fifteen years.

Instead, I'd recommend just graphing average usage over a year, a month, and a week periodicity and eyeballing it. You'll get better results with less work, despite using less tech.