Since several years, Cisco routers have included a feature called mpls autobandwidth that allows the router to compute automatically the bandwidth required for an LSP and adjust the reserved bandwidth. This feature seems interesting, but personally I have always feared that it could cause oscillations in a network if the change in reserved bandwidth leads to a change in paths for the concerned LSPs. I haven't seen papers that discuss or analyse this problem and show that a solution could be stable and that the autobandwidth feature could not lead to annoying cascade effects. A recent paper entitled MPLS automatic bandwidth allocation via adaptive hysteresis studies the bandwidth allocation problem. It analyses the bandwidth allocation technique used by cisco and proposes an alternative technique. The two techniques are evaluated based on mathematical models and simulations, but unfortunately the paper does not address the problem at the network level. A possible starting point for such a work could be the traffic matrices that we have collected on the GEANT network.