Routing In Fat Trees
charter-ietf-rift-01-02

Data Centers have been steadily growing to commonly host tens of thousands
of end points, or more, in a single network. Because of their topologies
(traditional and emerging), traffic patterns, need for fast restoration,
and for low human intervention, data center networks have a unique set of
requirements that is resulting in the design of routing solutions specific
to them. Clos and Fat-Tree topologies have gained popularity in data center
networks as a result of a trend towards centralized data center network
architectures that may deliver computation and storage services.

The Routing in Fat Trees (RIFT) protocol addresses the demands of routing in
Clos and Fat-Tree networks via a mixture of both link-state and distance-vector
techniques colloquially described as 'link-state towards the spine and distance
vector towards the leafs'. RIFT uses this hybrid approach to focus on networks
with regular topologies with a high degree of connectivity, a defined
directionality, and large scale.

The RIFT WG has finished the base protocol specification
(draft-ietf-rift-rift-20) and will continue to work on standards track
specifications on the following:

• Key-Value Data Store
• Policy Guided Prefix
• Segment Routing - RIFT extensions to support Segment Routing
• Using RIFT Zero Touch Provisioning procedures (as the management plane) to carry attributes that allow provision and instantiation of other protocols, such as:
  • EVPN, by distributing EVPN attributes (defined in RFC7342) such as VNIs, RTs and RDs
  • IS-IS, by distributing Flood Reflection attributes (defined in RFC9377) such as Flood Reflection Cluster IDs and related IS-IS attributes
• Multicast - RIFT extensions to allow building of multicast trees
• Leaf ring topologies - RIFT extensions to build leaf ring topologies
• Dragonfly topologies - RIFT extensions to build ToF ring topologies

The RIFT WG will also explore the use and extensions of the RIFT protocol for
the newer AI/ML Data Center architectures.