The paper uses TTL approximation to analyze the attack resilience of a number of state-of-the-art cache replacement policies (including several two-staged policies), in terms of how well they preserve the hit ratio for legitimate flows under a given number and rates of attack flows. Observing that no single policy is optimal under all kinds of attacks, it also proposes to adapt the policy based on coarse estimates of attack parameters, and demonstrates the efficacy of this approach based on real traces. This is Tian’s first paper with me. Congratulations, Tian!