Oh! So you think its that easy as 2 rusty cans and a slack string??

Maybe you possess the same views about the spam filtering softwares, the disease testing softwares, the Microsoft Office Assistant and maybe even the Autoclass project by NASA.

See for yourself:

http://ic-www.arc.nasa.gov/ic/projec...ass/index.html

Though I admit that Bayesian analysis comes with its own set of limitations.

The most striking problem is that of the aprior distributions and the belief that each of the factors is equally important.

And secondly, sec_ware fully illustrated it in his fourth point. Exploration of unknown networks through this frameworks may become costly in real time as it would involve calculation of probabilities of each branch and given the number of variables that can arise it would also become much more time consuming process.

Still its higly effective in case of a small network such as a LAN or even a WAN spanning some two or three blocks.