XBF: Scaling up Bloom-filter-based Source Routing

Abstract: A well known drawback of IP-multicast is that it requires per-group state to be stored in the routers. Bloom-filter based source-routed multicast remedies this problem by moving the state from the routers to the packets. However, a fixed sized Bloom-filter can only store a limited number of items before the false positive ratio grows too high implying scalability issues. Several proposals have tried to address these scalability issues in Bloom-filter forwarding. These proposals, however, unnecessarily increase the forwarding complexity. In this paper, we present Extensible-Bloom-filter (XBF), a new framing and forwarding solution which effectively circumvents the aforementioned drawbacks. XBF partitions a network into sub-networks that reflect the network topology and traffic patterns, and uses a separate fixed-length Bloom-filter in each of these. We formulate this partition assignment problem into a balanced edge partitioning problem, and evaluate it with simulations on realistic topologies. Our results show that XBF scales to very large networks with minimal overhead and completely eliminates the false-positives that have plagued the traditional Bloom-filter-based forwarding protocols. It furthermore integrates with SDN environments, making it highly suitable for deployments in off-the-shelf SDN-based networks.