What are the pieces needed to achieve full automation in the network?

In part one and two of our blog series, we’ve explored the need to have closed-loop automation per domain and the convergence of assurance and fulfillment systems. But what about artificial intelligence?

Recently, a buzzword has emerged in the telecom industry when describing the future of networking: machine learning. This isn't surprising as the makeup of the technology fits into the recipe for automation.

blur-1853305-656989-edited.jpg

Welcome to the machine

A subset of artificial intelligence, machine learning provides the ability for computers to learn through algorithms and perform tasks as they’re exposed to data. The more data that’s fed into it, the more a machine can build on its analysis, and can further adapt and solve complex problems without specific instructions.

The system used to detect credit card fraud is a notable example of this. The software would track your transaction tendencies and if any anomalies were to occur it would raise alarms.

While the technology’s functionality is already being tapped into by Amazon, Google, and PayPal, it has yet to receive the same level of interest from the telecom industry.

The theory of everything… sort of

There's no denying the significant opportunities machine learning presents, so why hasn’t it been applied to networking yet?

First, the notion that a machine can simply ingest all of the data coming from the network and figure out how to resolve problems isn't the whole story. While machine learning does have a role to play in automation thanks to its trending and analytic capabilities, it requires a solid foundation of data to be in place in order to be successful. This means establishing an accurate model of network and service topology to understand the dynamic attributes of underlying resources.

Second, machine learning is based on algorithms, and algorithms require models as building blocks. What networks lack that machine learning needs is a set of core principles. Without a theoretical model defining how all networks should be constructed and perform, it would require that a machine learn each network individually. This would result in escalating cost and complexity.

In order for machine learning to be utilized in predicting service impact and diagnosing faults in the network, it’s necessary to have a base of data to begin the analysis.

What's next?

What lies ahead for the future of service providers is unclear. However one thing is for certain, the complexity and scale offered by SDN and NFV have service providers re-evaluating how they manage their network.

In the age where customers rule, the industry is very business focused. The road to lights-out operations is entirely possible but it means larger investments over time as service providers build out systems capable of instantiating and assuring virtual network functions without the touch of a human. Are they willing to take the next step?

Download the Appledore Research Group’s white paper to learn how increased automation is the only way forward to support the demands of cloud services.

HybridNetworks.jpg

Subscribe Our Blog

HybridNetworks.jpg

Let Us Know What You Thought about this Post.

Put your Comment Below.