Load characterization using Markov Chain

Well, in this post I’d like to throw in the desk some of my last researches about markov chains. I’ve presented something seminal (as the overall state of the work at this moment) ’bout this article I’m working on here (in portuguese), where I’m trying to create a Markov Chain Model to represent (and afterwards simulate) a disk and processor scenario.

A Markov Chain (MC) is very simple to understand, and there’s a good introductory article in Wikipedia.

That’s easy to work with MC because of it’s mathematical formalism, thats easy to understand and well founded, thanks it’s matrix representation. Going straight at the point, all the hard work is to extract the probabilities of changing to a state to another, and in the case of disk analysis, and when you’ve metrics like % Disk time, %disk read time and % disk write time, it’s quite hard to know about the state of the disk in a moment (if it’s reading, writing or in idle state).

If someone got any idea, it’s all welcome. Until then, I’m going to try some statistical acrobacies to see if this problem can be solved.

See ya!

Other references ’bout Markov Processes and Models:

http://en.wikipedia.org/wiki/Hidden_Markov_model
http://en.wikipedia.org/wiki/Semi-Markov_process
http://en.wikipedia.org/wiki/Markov_process
Queueing Networks and Markov Chains – G. Bolch, S. Greiner, H. de Meer & K. S. Trivedi