Tradeoffs in Probabilistic Packet Marking for IP Traceback

Micah Adler
University of Massachusetts, Amherst

micah@cs.umass.edu

Thursday, October 4, Volen 101, 2:10-3:10 pm. (Refreshments at 2:00pm)

There has been considerable recent interest in probabilistic packet marking schemes for the problem of tracing a sequence of network packets back to an anonymous source. An important consideration for such schemes is b, the number of packet header bits that need to be allocated to the marking protocol. However, all previous schemes belong to a class of protocols for which b must be at least log(n), where n is the number of bits used to represent the path of the packets. In this talk, we introduce a new marking technique cing a sequence of packets sent along the same path. This new technique is effective even when b=1. In other words, the sequence of packets can be traced back to their source using only a single bit in the packet header. With this scheme, the number of packets required to reconstruct the path is O(2^(2n)), but we also show that Omega(2^n) packets are required for any protocol where b=1. We also study the tradeoff between b and the number of packets required. We provide a protocol and a lower bound that together demonstrate that for the optimal protocol, the number of packets required (roughly) increases exponentially with n, but decreases doubly exponentially with b. The protocol we introduce is simple enough to be useful in practice. We also study the case where the packets are sent along k different paths. For this case, we demonstrate that any protocol must use at least log(2k-2) header bits. We also provide a protocol that requires log(2k+1) header bits in some restricted scenarios (to which the lower bound applies). This protocol introduces a new coding technique that may be of independent interest.

Host: Marty Cohn