With smaller networks, the core/distribution/access layer topology is generally sufficient. This three-layer architecture is especially efficient in “North-South” data transfer, that is, traffic that comes into and leaves the Data Centre. The transmission latency between two internal points of the network is generally not a great concern.
However, in Enterprise and Campus networks with a Data Centre at the core, it can become an issue.
In the traditional architecture, the problem was addressed by increasing the number of connections between distribution nodes to create redundant pathways that could be used in the case of network outages or excessive congestion. This could create problems with network loops. Spanning Tree technology was developed to switch off redundant pathways and only activate them when needed. It also ensured that network latency was not predictable and manageable because the number of nodes a packet travelled though could vary. It also ensured that a good proportion, at least 50% and sometimes more of your available connectivity bandwidth sat unused.
Traditional networks, therefore, become increasingly complex and expensive and less and less scalable and manageable as they grew. As backbone speeds have increased, and the increased data transmission demands of multimedia applications become the norm, the effects of network congestion and route denial have shown that the traditional three-layer architecture can bring problems and needs a revamp.
The problem has been exacerbated by the deployment of Cloud architectures and the effects of digital convergence, especially the Internet of Things. A failure to process large volumes of transactions in the expected time can be business and sometimes life-threatening. A good example is automated vehicles. Several techniques have been developed to overcome this. Fog Computing sets up mini-clouds at the network edge where the transactions are processed. In effect only aggregated and necessary transactions come to the Data Centre core.
Most traffic in a Cloud environment tends to be “East-West”, that is between servers in the Data Centre. In an Enterprise Network designed for North-South traffic, having most traffic move East-West can create network congestion, unreliability in transfer times and wasted bandwidth.
A solution that has been proposed is a spine and leaf topology, a topology that ensures that all devices are the same number of network hops away, thereby providing predictable and consistent network latency. This is a precursor of the existing Software Access Defined networks, and of the future Intent Based networks.
Instead of three layers, the leaf and spine network has only two layers, the Leaf and the Spine. Access switches make up the Leaf layer. They are connected to devices like servers, firewalls, and edge routers. The spine layer is essentially a routing layer, the network backbone. Each Leaf switch is connected to each and every Spine switch.
This is an “East-West” architecture where data travels within the Data Centre. Spanning trees are no longer needed.
The advantages of a Leaf and Spine architecture include:
Layer Three Routing
Unlike with the Spanning Tree protocol, all routes are configured as active routes. Equal-Cost Multipathing (“ECMP”) makes it possible to use all the connections simultaneously while avoiding network loops and providing network stability. It is now possible to use the full available network bandwidth and not have at least 50% of it sitting unused waiting for a network failure.
In the past with Layer 2 routing and Spanning Trees, great care had to be taken to configure all devices correctly otherwise you could be left with a highly inefficient and possibly unstable network. Layer three routing removes that constraint, by allowing Spanning Tree Protocols to be removed.
A problem with the traditional three-layer architecture is link oversubscription, where a link is asked to handle more traffic than it can at any one time. This was normally rectified by adding additional links or switches to cope with the excess traffic. Sometimes this was difficult and certainly needed some network reconfiguration and downtime.
In a Leaf and Spine environment, the process is much simpler. Add a new Spine switch, connect it to each leaf switch, and the oversubscription is resolved. If the problem is a shortage of device ports on the leaf switch, simply add a new leaf switch and connect it to each Spine switch. In a software-defined access network environment, reconfiguration is automatic.
The use of a leaf and spine architecture increases the ability to take full advantage of a virtual server environment, particularly where mobility within the Data Centre is a requirement. The combination of Layer three routing and Software Defined Access network allows mobile servers to be supported anywhere in the Data Centre.
If you operate an Enterprise or Campus sized network and intend to make full use of a Software Defined Access network, or are struggling with managing an unstable network, the shift to a Leaf and Spine network will bring dividends. The transition is not easy but will prove worthwhile.