We present Elmo, a system that addresses the multicast scalability problem in multi-tenant datacenters. Modern cloud applications frequently exhibit one-to-many communication patterns and, at the same time, require sub-millisecond latencies and high throughput. IP multicast can achieve these requirements but has control- and data-plane scalability limitations that make it challenging to offer it as a service for hundreds of thousands of tenants, typical of cloud environments. Tenants, therefore, must rely on unicast-based approaches (e.g., application-layer or overlay-based) to support multicast in their applications, imposing bandwidth and end-host CPU overheads, with higher and unpredictable latencies. Elmo scales network multicast by taking advantage of emerging programmable switches and the unique characteristics of data-center networks; specifically, the hypervisor switches, symmetric topology, and short paths in a datacenter. Elmo encodes multicast group information inside packets themselves, reducing the need to store the same information in network switches. In a three-tier data-center topology with 27,000 hosts, Elmo supports a million multicast groups using an average packet-header size of 114 bytes (max. 325 bytes), requiring as few as 1,100 multicast group-table entries on average in leaf switches, and having a traffic overhead as low as 5% over ideal multicast.

We present Elmo, a system that addresses the multicast scalability problem in multi-tenant datacenters. Modern cloud applications frequently exhibit one-to-many communication patterns and, at the same time, require sub-millisecond latencies and high throughput. IP multicast can achieve these requirements but has control- and data-plane scalability limitations that make it challenging to offer it as a service for hundreds of thousands of tenants, typical of cloud environments. Tenants, therefore, must rely on unicast-based approaches (e.g., application-layer or overlay-based) to support multicast in their applications, imposing bandwidth and end-host CPU overheads, with higher and unpredictable latencies.

Most datacenters still use Equal Cost Multi-Path (ECMP), which performs congestion-oblivious hashing of flows over multiple paths, leading to an uneven distribution of traffic. Alternatives to ECMP come with deployment challenges, as they require either changing the tenant VM network stacks (e.g., MPTCP) or replacing all of the switches (e.g., CONGA). We argue that the hypervisor provides a unique point for implementing load-balancing algorithms that are easy to deploy, while still reacting quickly to congestion. We propose Clove, a scalable load-balancer that (i) runs entirely in the hypervisor, requiring no modifications to tenant VM networking stacks or physical switches, and (ii) works on any topology and adapts quickly to topology changes and traffic shifts. Clove relies on standard ECMP in physical switches, discovers paths using a novel traceroute mechanism, uses software-based flowlet-switching, and continuously learns congestion (or path utilization) state using standard switch features. It then manipulates packet-header fields in the hypervisor switch to direct traffic over less congested paths. Clove achieves 1.5 to 7 times smaller flow-completion times at 70% network load than other load-balancing algorithms that work with existing hardware. Clove also captures some 80% of the performance gain of best-of-breed hardware-based load-balancing algorithms like CONGA that require new equipment.

Multi-tenant datacenters predominantly use equal-cost multipath (ECMP) routing to distribute traffic over multiple network paths. However, ECMP static hashing causes unequal load-balancing and collisions, leading to low throughput and high latencies. Recently proposed alternatives for load-balancing perform better, but are impractical as they require either changing the tenant VM network stacks (e.g., MPTCP) or replacing all the network switches (e.g., CONGA).