L3 Leaf/Spine Networks and VXLAN

Sean Flack 
Consulting Engineer, EMEA
Arista Networks

Sean Flack - Arista - L3 leaf spine networks and VXLAN.pdf

For years, networks designs have looked very similar. There’s a core for connecting to the outside world, some form of distribution and an access layer for host/server connectivity. This has been driven by the physical restrictions and various media types that we have available or have to support. With practically ubiquitous support of Ethernet and IP and the advance of high density merchant silicon, large networks can be built using relatively low cost and small form factor devices.

To ensure linear scalability and modular growth without impacting network performance, subscription and latency, many network designs are moving to the leaf spine topology. The leaf spine architecture consists of a series of leaf switches connecting to a spine layer, that acts as a high performance switching fabric for interconnecting the various leaf switches. With this model all leaf switches and their associated hosts can communicate with every other leaf within the architecture with the same consistent latency and engineered subscription ratio, regardless of the traffic flow. Thus facilitating the dynamic placement of network services, compute and storage anywhere within the architecture while maintaining a consistent network performance.

Today, network switches have dozens of 40G ports available in a single rack unit. Using such devices in leaf spine topologies means the reliance on large, power hungry, complicated modular devices is gone. This also means networks can become more stable, more scalable and cheaper to purchase and run. As networks grow, higher capacity devices can be introduced into the network with minimal impact and avoiding a entire overhaul of the network.

One challenge with leaf spine networks is the ability to transport traffic within the same subnet across physical locations. VXLAN overcomes this by allowing the provisioning of L2 domains, called virtual network identifiers (VNIs), over a standard L3 architecture. This allows resources and hosts to be L2 adjacent regardless of their physical location.

This talk will explain some of the benefits of leaf spine over legacy designs and some of the cost saving that can be realised. It will also cover some of the implementation details and challenges and how these can be overcome.

All talks:

If you require information about any aspect of the conference please contact: Fi Coyle: fiona.coyle@heanet.ie / +353 1 6609040