BloomCasting for publish/subscribe networks

Publish/subscribe has been proposed as a way of addressing information as the primary named entity in the network. In this thesis, we develop and explore a network architecture based on publish/subscribe primitives, based on our work on PSIRP project. Our work is divided into two areas: rendezvous and Bloomcasting, i.e. fast Bloom filter-based forwarding architecture for source-specific multicast. Taken together these are combined as a publish/subscribe architecture, where publisher and subscriber matching is done by the rendezvous and Bloom filter-based forwarding fabric is used for multicasting the published content. Our work on the inter-domain rendezvous shows that a combination of policy routing at edges and an overlay based on hierarchical distributed hash tables can overcome problems related to incremental deployment while keeping the stretch of queries small and that it can solve some policy related problems that arise from using distributed hash tables in inter-domain setting. Bloom filters can cause false positives. We show that false positives can cause network anomalies, when Bloom filters are used for packet forwarding. We found three such anomalies: packet storms, packet loops, and flow duplication. They can severely disrupt the network infrastructure and be used for denial-of-service attacks against the network or target services. These security and reliability problems can be solved by using the combination of three techniques. Cryptographically computed edge pair-labels ensure that an attacker cannot construct Bloom filter-based path identifiers for chosen path. Varying the Bloom filter parameters locally at each router prevents packet storms and using bit permutations on the Bloom filter locally at each router prevent accidental and malicious loops and flow duplications.