I recently visited a friend of mine in Paris who is now working for a N2NSoft, a company with quite an interesting product: Before deploying networks or even before applying for a license, operators need to know how much to spend on the wireless and fixed infrastructure to see if their business model is sound. The cost and number of base stations, routers, fiber links, etc. mainly depends on the number of users, what kind of applications their are using, if they are mobile or not, how good the coverage of the network is, etc. etc. As there are so many variables, simple calculations might give a ballpark estimation of the costs to come. However, this is not good enough for financial calculations.
If one could simulate the user behavior, however, a prediction could be much more accurate. This is just what the company has set out to do with their flagship product called NetScale. Simulating the behavior of a couple of users is simple in terms of computing power. So far, however, processing time of most algorithms used grow exponentially with the number of simulated users. So simulating more than a few users gets tricky very quickly. It looks like N2NSoft has found a solution to this with an algorithm developed at the INRIA, one of the top French research institutions. Instead of computing time growing exponentially, their algorithm scales linearly. This is the dream of any mathematician.
Over lunch, we played with a network simulation of 250 nodes and 50.000 users. Each user was simulated individually and the simulated network had all kinds of different users from stationary notebook usage to moving hand-held eMail retrieval and ring-tone downloads. For each user, the simulation shows the TCP/IP traffic flow in real time, packet losses, retransmissions and overall performance on every link. To see the effect on end user throughput when capacity on a link is reduced or increased is simple as everything is computed in real time. Click on the picture on the upper left, it shows a bit of the GUI and simulation output. More pictures and info can be found here.
Thanks for the interesting lunch break, Pierre, I keep dreaming at night of 50.000 mobiles all accessing the network simultaneously 😉