A GSM routing experiment

Once, a few years ago, me and two of my colleagues (I'll call us A,B,C) have made on purpose a circular call diversion on our GSM phones: A -> B -> C -> A. We were interested in what will happen if any of us gets called. All of us could normally call anybody without a problem so we called a 4th colleague to call one of us from a fixed line. The result - an automatic machine saying that the person is temporarily unavailable.

Since we were allowed to make circular diversions in the first place, I think that there does not exist a central database of all diversions. They are executed only when needed. So, how smart are these GSM switches in loop detection, and what are their limits?

Consider the same experiment, extended in several ways:

1. There also exists an 'unavailable' diversion which takes place only if the person is, well, not available. This makes it possible to create more complicated graph topologies than simple closed rings. The more complicated the topology - the longer it takes to find a 'leaf' node by DFS (one not involved in any cycle).
2. Invent a suitably complicated topology.
   
3. Involve more, many more, people. On the order of few hundreds to few thousands. Instruct them how to arrange their call diversions. Explain to them that the experiment will cost themnothing.
   
4. Get those people spatially close, in groups of about 40. I think that a single base station cannot handle more than about 30 users. So all of them will (maybe) overload a single base station. Depending on the software.
5. Maybe involve people from different countires and/or providers in the same country. They can't share data about the (un)availability of a user.
6. Perform a call from the fixed line to a well-chosen user in the arranged 'diversion graph' and observe what happens.

I can envision several outcomes:

• The software simply does not respect the 'not available' diversion if it detects a loop. This leaves us with a simple ring topology and nothing interesting will happen. The caller will simply get the 'user not available' message.
• The software has a maximum limit on the depth of DFS search for the final divert destination. So at some point, the cycle will be artificially broken by the built-in limitation and the phone of some user that is in the cycle will ring.
• The software is really smart and traverses the whole graph before concluding that it is cyclic and replying by 'user not availabe'. A lot of the operator's bandwidth could be consumed in the process, maybe temporarily disrupting the service.
  
• The operator service totally breaks down, either due to software failure or due to the overloading of base stations (too busy communicating trying to find the end user and not servicing 'real' calls).

If this is read by some engineer who has designed a GSM switch and/or software... please, do not hesitate to leave comment about its inner workings!

On the other hand, if someone seriously decides to make real the experiment I have just described.. I volounteer to invent an 'interesting' topology :)