Improving Data Center Multicast with In-Packet Lightweight Bloom Filter

IP Multicast benefits data center networks (DCNs) for saving bandwidth and improving throughput. In-packet Bloom filter (BF) has been recently proposed to achieve efficient and scalable multicast in DCNs, where a Bloom filter encodes all nodes in a multicast tree into the packet header for making routing decisions. However, prior work incurs large traffic overhead due to tree encoding and false positive forwarding. In this paper, we propose an improved data center multicast scheme called in-packet lightweight Bloom filter (LBF) for significantly reducing traffic overhead. The basic idea behind LBF is to encode the switches of a multicast tree into the Bloom filter carried in the packet header, which facilitates decreasing the tree encoding length and false positive rate. Each forwarding switch examines the in-packet LBF for multicast packet forwarding, while each edge switch in the tree uses a hash table lookup for forwarding packets to the receivers on the servers. Simulations and experiments on Click prototype show that in-packet LBF reduces the traffic leakage ratio by up to 39.3% as well as the tree encoding length by 38.8%, and improves the packet forwarding rate by up to 98.7% compared to in-packet BF.