Can TCP metastability explain cascading failures and justify flow admission control in the Internet?

This paper discusses implications of possible metastability of TCP-type fair bandwidth sharing under random flow arrivals/departures for understanding and defending the Internet against cascading failures. Cascading failures can be viewed as a process of network transition from desirable metastable mode with finite number of flows in progress to the congested mode. It is possible to eliminate or reduce the possibility of cascading failures with properly designed flow admission control which stabilizes the network in a close neighborhood of the desirable metastable state.