Dust particles glittered in the shaft of light that sliced across the room. Floating free they lazily wafted through the air, following a randomly aimless journey. The sun was warm on my face; lunch was heavy in my stomach. Dave was doing his best to impart some knowledge of network technologies to the assembled group. It was a Friday afternoon and Time Division Multiplexing, despite Dave’s best efforts, was inevitably losing the battle of wills against the thought of the Ocean Youth Club party in Hamble on Saturday night.
I think the low point was the OSI 7 layer model, which seemed a particularly academic exercise with little practical value, especially as it had been developed some years after the deployment of data networks and had been retrofitted to what was already being used for real. Every second that afternoon seemed like a minute, every minute an hour, as time became as viscous as treacle.
Several years after this training session – more years later than seems possible with my elder son now starting his BA (Hons) in Sculpture – those faint memories of the OSI model need to be dug out of the bottom draw and given a good shaking in the fresh light of a new day. It’s time for some revision!
The OSI 7 layer model describes networks as a hierarchical stack, with various network functions neatly compartmentalised into each layer. For example, SDH sits at Layer 1 of the OSI model and is often used for point-to-point leased lines.
The traditional approach to delivering a multi-site solution is to build an overlay network of Frame Relay on top of the SDH network. Frame Relay sits at Layer 2 of the OSI model and provides point-to-multipoint connectivity. It doesn’t have its own Layer 1 capability so has to use the only game in town – SDH.
It’s the same story for ATM, but as ATM is a “fully meshed” multi- site solution the problems caused by the overlay network approach are amplified. The number of Layer 1 SDH circuits needed to support a Layer 2 overlay ATM network is given by the formula: n x (n-1) where n is the number of customer sites. This becomes very expensive for the customer.
But hasn’t the world “converged” on IP networks? Well yes, and no. Look at the model again – IP is Layer 3 and TCP Layer 4 – neither can do anything without Layers 1 and 2 in place, resulting in a “fully stacked” hierarchy of overlay networks.
A “converged IP network” is one that uses IP as a standard addressing format and standard packaging format for content. The rest of the network is made up of a mix of SDH-based Frame Relay at the WAN edge, SDH-based ATM in the WAN core and Ethernet in the LAN.
The resultant hierarchical stack of overlay technologies in the WAN is expensive to build and therefore costly for the customer to use. It’s complicated, inflexible in terms of bandwidth and doesn’t deliver the performance necessary for applications such as Voice Over IP and Storage.
By contrast, Optical Ethernet networks are “fully de-layered” and deliver the network attributes that customers seek. Ethernet straddles both Layer 1 and Layer 2 of the OSI model – it has its own Layer 1 component built in (known as PHY), so it doesn’t need a separate Layer 1 technology such as SDH to provide it.
This is why SDH is not used in the LAN and why running Ethernet (Layer 1 and Layer 2) over SDH (Layer 1) in the WAN is pointless. Not only does SDH duplicate the Layer 1 function that Ethernet can do just fine on its own, it also adds unnecessary complexity, rigidity and expense to the network.
But Ethernet’s a Local Area Network technology, right? Isn’t SDH needed to get the Ethernet around the country? Not necessarily so. MPLS, and in particular the Martini draft of MPLS deployed at Layer 2, overcomes the few limitations of Ethernet as deployed in traditional LANs. The recommended “7 hop limit” no longer applies and Spanning Tree is no longer required.
The 7 hop limit was recommended so that network convergence times would be kept reasonable – if the limit was exceeded a LAN would take minutes, or would even fail, to re-converge (ie to re-establish a failed link) with a consequent detrimental effect on network performance. From a service provider’s perspective, 7 hops won’t allow a network with national reach to be built, hence the restriction of “classic” Ethernet to the LAN.
Spanning Tree was a great way to ensure data didn’t end up going uselessly around and around in loops, however it also wasted bandwidth. MPLS optimises network design by pre- configuring the primary and failover routes, thus avoiding loops and wasting of bandwidth.
In traditional Layer 3 IP networks, OSPF (Open Shortest Path First), and other similar routing protocols, will always select the shortest route and ignore better or faster routes. This results in network congestion. MPLS at Layer 2 overcomes this by allowing us to pre-select or pre-configure the best routes, rather than the shortest routes, through the network and to remove this source of network congestion. In Neosnettm latencies are down to sub- 10ms on the national network and failover times are sub-50ms.
Neos has the first operational deployment of Nortel’s Optical Gigabit Ethernet long haul cards. These cards drive the optical signal over enough distance for us to economically create a national, de-layered, optical Ethernet network. This enables us to deploy Ethernet directly over DWDM around the UK without resorting to Ethernet as an overlay to SDH.
We have created and launched point-to-point, and point-to- multipoint Ethernet VPNs by merely changing software parameters in the Layer 2 Ethernet switches. We have not had to build a separate Layer 2 point-to-multipoint overlay network. We have therefore not only displaced Layer 1 SDH but have also displaced the Layer 2 overlay networks of both Frame Relay and ATM.
So I guess that OSI 7 layer model was relevant after all. Thanks, Dave.