Make best-effort forwarding upon network abnormality

Abstract

The current routing system of the Internet is in nature best-effort. We often observe, however, that the Internet routing today drops packets upon network abnormalities; even when the network is not congested, and/or physical paths exist. Such strategy usually makes the routing design simple; and even the packet is dropped, well, we know we have TCP. Nevertheless, the key function of any routing system, by its definition, is not to drop a packet, but to deliver. In this paper, we rethink this strategy in the BGP convergence period after an abnormality happens. During such period, the network states are inconsistent. As a result, the ASes may not have outgoing AS paths for the packets, or the packets may travel in an AS-level loop until their TTLs are exhausted. Our key idea is that rather than dropping the packets, the ASes will make a best-effort decision to forward the packets, sometimes even by random. We first verify our idea through a real Internet experiment using CNGI-6IX, AS 23***. We observe that the current Internet infrastructure is able to deliver a packet; when an AS forwards this packet to a neighbor AS that is not the original next hop of this packet. We then proposed two simple best-effort forwarding schemes. Our design requires minimum change in the current hardware and is incrementally deployable. Our preliminary simulation also shows that 40%-70% of packets that deem to be dropped are delivered using our scheme.