IP fabric Infrastructure

IP Fabric

An IP fabric is one of the most flexible and scalable data center solutions available. Because an IP fabric operates strictly using Layer 3, there are no proprietary features or protocols being used, so this solution works very well with data centers that must accommodate multiple vendors. Some of the most complicated tasks in building an IP fabric are assigning all the details, like IP addresses, BGP autonomous system numbers (AS numbers), routing policy, loopback address assignments, and many other implementation details. 

Three Stage IP Fabric Architecture

In the 1950s, Charles Clos first wrote about his idea of a nonblocking, multistage, telephone switching architecture that would enable calls to be completed. The switches in his topology are called crossbar switches. A Clos network is based on a three-stage architecture — an ingress stage, a middle stage, and an egress stage. The theory is that there are multiple paths for a call to be switched through the network, such that calls can always connect and not be "blocked" by another call. The term Clos “fabric” emerged later because of the pattern of links resembling threads in a woven piece of cloth.

 In an IP fabric, the ingress stage and the egress stage crossbar switches are called leaf nodes. The middle-stage crossbar switches are called spine nodes. In a spine-and-leaf architecture, the goal is to share traffic loads over multiple paths through the fabric. A three-stage fabric design ensures that the access-facing port of any leaf node is exactly two hops from any other access-facing port on another leaf node. It is called a three-stage fabric because the forwarding path from any connected host is leaf-spine-leaf, or three stages, regardless of where the destination host connects to the fabric

IP Fabric Design Options 

Currently, there are three prominent design options in an IP fabric architecture.

Option 1 is a single-stage fabric called a spineless fabric. A spineless fabric is usually deployed in a smaller network. Such a topology enables a data center to reap the benefits of an Ethernet VPN–Virtual Extensible LAN (EVPN-VXLAN) overlay (Ethernet segment identifier link aggregation [ESI LAG] and more) with no spine layer. Here, you see that there is a full mesh of connectivity (and BGP signaling) between each of the leaf nodes.

Option 2 is a basic three-stage architecture. Each leaf connects to every spine. The number of spines is determined by the number of leaf nodes, and the throughput capacity required for leaf-to-leaf connectivity. The diagram shows a spine-and-leaf topology with two spine devices and four leaf nodes. A single link from each leaf node to each spine node limits the throughput capacity to twice the capacity of a single uplink. A spine device failure would cut the forwarding capacity from leaf to leaf in half, which could lead to traffic congestion. You can increase this by adding additional uplinks to the spine nodes using technologies such as LAG. However, this type of design places a large amount of traffic on few paths through the fabric. To increase scale, and to reduce the impact of a single spine device failure in the fabric domain, you can add additional spine devices to the topology. If, for instance, you added two more spine nodes to Option 1, the traffic from Leaf 1 to Leaf 4 would have four equal-cost paths for traffic sharing. A spine device failure or maintenance that requires a spine device to be removed from the fabric would reduce the forwarding capacity of the fabric by one-fourth instead of one half.

---