Trellis is the leading open-source SDN based, multi-purpose L2/L3 spine-leaf switching fabric for data-center (DC) networking.

Leveraging the ONOS Controller, Trellis creates a non-blocking fabric for data centers using white box switching hardware and open source software. Unlike traditional networking approaches, the fabric itself does not run a control protocol (such as BGP, OSPF or RSTP). Instead, all the intelligence is moved into applications running on the clustered ONOS controller. In this way the fabric switches can be simplified, the entire fabric can be optimized by leveraging a holistic view of all activity, and new features and functionality can be deployed without upgrading the switches.

Trellis can optionally provide both an underlay network and multiple tenant overlay networks. The underlay connects all the physical servers and switches, while the overlays create private networks for separate users, applications or services running on top of the underlay. Trellis also provides an industry best-of-breed vRouter solution for external connectivity, where the entire fabric acts as a distributed virtual router data-plane.

Trellis is an integral part of the CORD platform, providing the network infrastructure for all connectivity in the CORD POD. Trellis can also be used independent of CORD for general purpose data center networking needs.

Traditional ways of building spine-leaf fabrics rely on control protocols running in each of the switch fabric nodes to create the network underlay. Several layer 2 or layer 3 control protocols (like BGP) are used, requiring each of the fabric nodes to implement and run one of these complex protocols. As a result, switches need to have more powerful CPUs, memory and all the complexity of this software software. This adds cost and vulnerability to failure.

Furthermore, overlays are then constructed on top of the underlay to create private tenant domains. These overlays are entirely independent of the underlay, eliminating the ability to coordinate and optimize overlays for change in the underlay. This construction requires hairpinning tenant traffic through virtual (software) Routers in VMs, that act as gateways between the virtual and physical networks.

Trellis addresses these challenges, and in so doing:

  • Simplifies the switching nodes (eliminating the need for complex control and routing protocols in the switches)
  • Creates a single routing instance for the entire fabric, simplifying how the fabric connects to the external network
  • Provides distributed virtual routing to all tenant traffic in the overlay, so their traffic can directly reach the physical network without having to hairpin to a gateway VM
  • Coordinates the underlay and overlay to optimize resource deployment and connectivity
  • Uses ECMP routing for multi-pathing and segment routing principles to offer fine grained path control for select traffic
  • Fully integrates the software switches running on each server, to provide a complete solution for interconnecting servers, applications, VMs and containers with dynamically created tenant domains and service-chains

Unified SDN Control of Leaf-Spine Fabric Underlay and Virtual Network Overlay

  • Provides all the connectivity between the access and compute-nodes
  • Allows the entire CORD POD to be viewed as a single BGP-speaking router to the external network
  • Optimizes all internal communication paths, while simplifying connections to the rest of the operator network

Trellis: Enabling Network Infrastructure for CORD

  • Eliminates complex control protocols in the fabric nodes
  • Reduces costs
  • Reduces risk of software failure in nodes
  • Integrates both underlay and overlay configuration and control
  • Provides a holistic optimized environment, providing the best application performance
  • Runs the entire fabric as an ‘SDN island’, presenting this island as a single BGP instance to the outside world
  • Simplified management and minimises the number of BGP instances in the network, thus speeding route recalculation times and minimizing BGP processing load throughout the network.
Trellis Architecture