Thursday, May 16, 2013

"Mix and Match" is one Advantage of Software Defined Networks


If you wanted to rip cost out of any device, app or network, what are some of the things you’d do? You’d remove parts. You’d probably reduce size.

Shoretel, for example, sells docking stations for iPhones and iPads, for example, that allow the iOS devices to act as the CPU, when the docking station providing all the peripheral support.



And that's as good an example as any of how a network designer would try and wring cost out of a network.


You’d rely as much as possible on functionality that could be supplied by other common CPUs. 

You’d pick passive solutions, not active solutions, as often as possible. You’d simplify with an eye to improving manufacturing tasks.

You also would create a “mix and match” capability about specific makes and models of network gear. You’d want to be able to use a variety of network elements, made by many different suppliers, interchangeably.

You’d create a network using common, commodity-priced devices as much as possible.

In other words, you would make devices, networks and apps "as dumb as possible," and as flexible and interchangeable as possible.

If you think about software defined networks, that’s an application of the principle. Not “everything” about SDN is “as dumb as possible;” only the data plane elements.

The corollaries are that such approaches create networks that also are “as flexible as possible” and “as affordable as possible.”

The control plane you would still want to be as “smart as possible,” and you would be able to afford to do so, since the key to much cost reduction is the ability to share a physical resource across a broad number of end users, subtending devices or nodes.

That is why the per-user or per-customer cost of an expensive headend is rather low, as a percentage of the total cost of a network. On the other hand, the total cost of smart CPUs (decoders) used by every end user or customer is so high because there is so little sharing possible: each customer needs one or more decoders.

That was what drove cable operator Cablevision Systems to adopt a “network digital video recorder” strategy. By centralizing the CPU functions, the ability to share the cost of processing infrastructure was vastly improved.

The broader principle is that one proven way to reduce cost, increase flexibility and enhance functionality is to separate and centralize control plane (CPU) functions from the data plane functions that are widely distributed throughout a network.

That’s the whole point of software defined networks.


No comments:

Costs of Creating Machine Learning Models is Up Sharply

With the caveat that we must be careful about making linear extrapolations into the future, training costs of state-of-the-art AI models hav...