[Lars Fischer gave an interesting presentation today on the challenges of deploying and managing federated network services. I think there is general agreement that R&E networks are moving away from hierarchical structures as have today with GEANT, Internet 2, CANARIE etc to a more heterogeneous mix of networking solutions including hierarchical, federate and virtual resources. This is being driven by several factors: (a) the interest of regional networks to directly peer with each other, (b) the localization of Internet traffic due to content peering, (c) the availability of cross border fiber, etc
As an alternative to hierarchical networks, in his presentation Lars explores the deployment of Federated Networks. As he points out federations of any kind (not necessarily networking) involve lots of challenges in terms of coordination of NOCs, trouble shooting, management etc. This was the model used by the old PTTs and is still reflected in the architecture of telephone systems to this very day.
An alternative approach to Federated networks is to deploy virtual networks that use virtual routers, virtual network links and other virtual resources that span multiple independently managed infrastructure domains. This is the essence of the Manticore and UCLP projects. The advantage of virtual networks is that there is one single management domain for all network services and trouble shooting. Virtual networks can also be deployed on a fine grain basis for specific communities of interest such as high energy physics, astronomy etc.
In reality there is no true 100% virtual network. At some point in the network stack a virtual network operator has to negotiate and interact with multiple independent management domains- and so at some level some sort of federation is required. However if the federation is kept to lowest levels of the network stack such as assignment of wavelengths to virtual network operators then the virtual operator network can deploy a multi-path redundant network – an Articulated Private Network (APN) which requires a minimal interaction between separately managed and independent physical infrastructure networks. If there is an outage in one of the underlying domains the virtual network operator can quickly re-route on its other virtual links. Manticore has done a lot of excellent work in this area.
The other advantage of APNs is that they allow increased competition, as well as specialized network services and provide a choice to institutions and researcher of possible network solutions. There is no reason for example why one NREN could not extend its network into another NRENs territory to offer services to partner institutions or satellite campuses. This concept may be anathema to some NRENs who think of themselves as exclusive monopoly providers. But as long as the underlying infrastructure is funded through membership fees it would enhance the potential service offering for many NRENS.
As NRENs start to look to deploy virtual wireless 5G networks integrated with their optical backbones the debate of federated versus virtual will become more intense. Building a 5G Wifi network out of hundreds of federated campus WiFi networks will be extremely difficult, whereas a virtual Wifi, 3G/4G and optical network could be built from a mix of heterogeneous resources both commercial and academic and be more easily integrated with existing Mobile Virtual Network Services.
Lars Fischer’s Presentation