Thursday, August 9, 2018

Network Slicing Returns to Some Elements of Debate Over "Stupid or Smart" Networks

Network slicing, an attribute of virtualized core networks, is in many ways the latest twist in thinking about where communications network intelligence should reside. Significantly, network slicing also is foundational for thinking about what can be done with 5G networks, and why 5G platforms will be different from all earlier mobile generations.

What one can do in the core of the network fundamentally drives what services can be created and offered at the edge of the network, of course. But in the IP era, that has been complemented by a rival approach, namely creating services and value at the edge of the network.

The best examples are any “internet” apps that are created by a combination of end user demand at the edge (from personal devices) and supply by data centers likewise at another edge of the network. In other words, cloud computing is the best example of value, apps and services created entirely at the edges of any communications network, and not in the core of the network, by the network itself.

But some changes are coming, at least for communications network operators. As core networks are virtualized, and as 5G gets deployed, it will be possible to create, as a principal network function, virtual networks that can be optimized on a number of dimensions to provide differentiated features.

Quality of service, latency performance, geographic reach, security, degrees of mobility and other features of the network service can be optimized for particular use cases. The shift is from one network with a standard set of capabilities to essentially multiple networks optimized, to an extent, for the use cases.

One use case might lean heavily on latency performance; another might require predictable packet arrival; throughput or security. The promise of network slicing is that, in principle, custom networks can be created.

Some of you, with long memories, can remember heated debates in the late 1990s and early 2000s about the “right” architecture for communications networks.

Telecom professionals argued for “smart networks” with intelligence in the core, while those in the data communications camp argued for “dumb networks” with intelligence at the edge.

Some of you will remember this as the debate over whether asynchronous transfer mode (ATM) or IP should be the protocol used by the next generation network. Perhaps the best example is the argument made by David Isen that the future network should be a stupid network, putting intelligence at the edge, as IP devices do.

That sort of thinking got a heated response. But, in a fateful decision, everybody decided IP was the better choice, as a practical matter. No matter the invective, the “stupid network” became the preferred network for data, even on “telco” networks.

But the basic debate never dies down completely. New thinking could arise, at least for telecom architects, as networks are virtualized. In principle, that is a move in the direction back towards core network intelligence.

One might argue that all that is involved with NFV or software defined network approaches is where intelligence is situated (in fewer network elements) or who supplies those elements (specialized or general purpose; branded gear or white box platforms).

But there always are other dimensions. Firms that specialize in supplying IP bandwidth often fall in the “dumb network” camp, by definition: their product is, in fact, simple IP bandwidth.

"I'm much more in the 'not only dumb pipe, but dumb network' camp," said Dave Schaeffer, CEO of Cogent Communications Holdings.

Virtualized networks needed to support 5G, on the other hand, supporting network slicing, for example, shift back towards smart networks, by definition, as the network itself has to act to create virtual private networks as a core function. Up to this point, VPNs have functioned as “higher in the stack” overlays on IP transport.

Network slicing makes creation of virtual networks a core network capability. It is not exactly a return to the completely “smart” or “intelligence in the core” philosophy, but it is a significant step in that direction, without the overhead and complexity of legacy telecom networks, one might hope and expect.

No comments:

"5G or 4G" Will Not Matter for Most Users, in Practice

It already is tough to determine what “5G speed” actually means, as early 5G often is based as much on 4G as 5G. And the problem of a...