A Heuristic Algorithm for Relaxed Optimal Rule Ordering Problem

The packet classification problem aims to determine the behavior of incoming packets at network devices. The linear search classification algorithm assigns each packet according to its prior actions, which are determined by comparing the packet header with classification rules until a match is found. As the processing latency of packet classification is proportional to the number of rules, a large number of rules can result in serious communication delay. This problem is generalized to Optimal Rule Ordering (ORO), which aims to identify the rule ordering that minimizes the delay caused by packet classification. If two different rules match a single packet, conventional ORO does not allow the posterior rule to be placed in a higher position than the prior rule. However, interchanging the rules does not violate the policy if the actions of those rules are the same. Thus, in this paper, we specifically consider the Relaxed Optimal Rule Ordering (RORO) problem, in which rules can be interchanged if their actions are the same. In RORO, the weight of rules may vary as they are interchanged. Hence, we propose a method of calculating the weights using a zero-suppressed binary decision diagram. We prove the difficulty of estimating the weights and propose an algorithm for RORO. This algorithm computes a rule list that ensures lower latency than in several conventional algorithms and accurately computes the latency. We demonstrate the effectiveness of our method by comparing it with previous models and reordering methods.