Ernie: Data Center Multicast Source Routing

Abstract

Multicast offers a prime means to support group communication in data center (DC) networks, as it saves network traffic and improves application throughput. Unfortunately, traditional IP multicast routing does not scale well with the number of supported multicast groups when used for cloud DC networks. Prior attempts are shown to scale multicast by taking advantage of the symmetry of DC topologies and the programmability of DC switches to compactly encode multicast group information inside packets. However, when considering large multicast group sizes, these approaches incur high network and/or CPU overheads, and they struggle to balance between network overhead (i.e., switch memory and packet size overheads) and end-host CPU overhead. For instance, in these approaches, keeping network overhead minimal leads to inflating end-host CPU load and vice versa. This paper proposes Ernie, a source-routed multicast routing approach that addresses the multicast scalability limitations while keeping both the network and end-host overheads at low levels. Ernie further exploits DC network structural properties and switch programmability capabilities to encode and organize multicast group information inside packets in a way that minimizes downstream header sizes significantly, thereby reducing overall network traffic.