Reducing packet delay through filter merging

The use of packet filters has increased considerably due to the growth of Internet users and network services. A number of header fields must be examined by the filter, causing delay for each packet processed. The problem is compounded when considering multiple filters across a network. To maximize network performance, it would be desirable to minimize the number of packet filters for each path across a domain. Due to the interactions of rules between filters, the underlying network topology and the actions of dynamic routing protocols, it is computationally infeasible to implement this strategy by collectively considering all packet filters across the network. A simpler approach is the elimination of a filter by merging two filters on a common network segment. This work presents a novel packet filter merging algorithm using decision diagrams. A large number of practical and simulated experimental results are provided to demonstrate the effectiveness of the technique and possible enhancements are considered in the conclusion. The results show an average 20% performance improvement can be obtained using the technique.

The use of packet filters has increased considerably due to the
growth of Internet users and network services. A number of header
fields must be examined by the filter, causing delay for each packet
processed. The problem is compounded when considering multiple
filters across a network. To maximize network performance, it
would be desirable to minimize the number of packet filters for
each path across a domain. Due to the interactions of rules between
filters, the underlying network topology and the actions of dynamic
routing protocols, it is computationally infeasible to implement this
strategy by collectively considering all packet filters across the
network. A simpler approach is the elimination of a filter by
merging two filters on a common network segment. This work
presents a novel packet filter merging algorithm using decision
diagrams. A large number of practical and simulated experimental
results are provided to demonstrate the effectiveness of the
technique and possible enhancements are considered in the
conclusion. The results show an average 20% performance
improvement can be obtained using the technique.