Segment-Encoded Explicit Trees (SEETs) for Stateless Multicast: P4-Based Implementation and Performance Study

Abstract

IP multicast (IPMC) is used to efficiently distribute one-to-many traffic within networks. It requires per-group state in core nodes and results in large signaling overhead when multicast groups change. Bit Index Explicit Replication (BIER) and its tree engineering variant BIER-TE have been introduced as a stateless transport mechanism for IPMC. To utilize BIER or BIER-TE in a large domain, domains need to be subdivided into smaller sets of receivers or smaller connected subdomains, respectively. Sending traffic to receivers in different sets or subdomains necessarily implies sending multiple packets. While efficient algorithms exist to compute sets for BIER, algorithms for computing BIER-TE subdomains are still missing. In this paper, we present a novel stateless tree encoding mechanism called Segment-Encoded Explicit Tree (SEET). It encodes an explicit multicast distribution tree within a packet header so that tree engineering is supported and sets or subdomains are not needed for large domains. SEET is designed to be implementable on low-cost switching ASICs which we underline by a prototype for the Intel Tofino™. If explicit distribution trees are too large to be accommodated within a single header, multiple packets with different distribution trees are sent. For this purpose, we suggest an effective optimization heuristic. A comprehensive study compares the number of sent packets and resulting overall traffic for SEET and BIER in large domains. In our experiments, SEET outperforms BIER even for large multicast groups with up to 1024 receivers.

Keywords

• Segment-Encoded Explicit Trees (SEETs)
• bit index explicit replication (BIER)
• multicast
• IP networks
• performance evaluation
• optimization