How Far Can AT&T Go With Automation?

As their networks get bigger and more complex, service providers are finding that more automation rather than more staff is key for keeping everything humming along. There's no shortage of automation technologies to choose from, including artificial intelligence (AI), machine learning (ML) and robotic process automation.

Telco Transformation recently spoke with Mazin Gilbert, vice president of AT&T Labs , Advanced Technology and Architecture, about what his company has learned from how it’s implemented automation so far and what's next.

Telco Transformation: Service providers have to automate more network processes because it’s not cost-effective to have humans calling all the shots. What are some processes they could or should automate? And why focus on those?

Mazin Gilbert: AT&T has been automating network processes for many years. The challenge has been flexibility to automate management and orchestration of workloads, and life cycle management of network functions. However, software-defined networking and network function virtualization better enable us to do those things.

We began this journey at AT&T five years ago when we developed ECOMP, a platform for managing and orchestrating virtual and physical network functions. (See AT&T's Gilbert: ECOMP Is a Game Changer.) In 2016, we surpassed 34% virtualization of our target network. Thus far, we have deployed hundreds of virtual network functions -- running on our network cloud while managed and orchestrated by ECOMP.

Although the majority of these functions are for mobile, there are also a number of wireline network functions that we have deployed to take advantage of ECOMP. Our goal is to virtualize 75% of our target network by 2020. (See Donovan: AT&T Closes In on Tipping Point With Virtualization.) [Open source ECOMP was merged by the Linux Foundation with OPEN-O earlier this year to form ONAP. Gilbert is the technical steering committee chair for ONAP.]

TT: At some service providers, one barrier to wider automation is that they simply don’t trust AI, machine learning, robotic process automation, etc. So they keep a lot of humans in the loop to make the final call. A related concern is that the fewer humans in the loop, the more opportunities there are for hackers to exploit automation. Is that a valid concern? And what can service providers and/or their vendors do to minimize automation-related vulnerabilities?

MG: Most automation-related vulnerabilities are not caused by automation itself, but by the move toward software. At AT&T, we are building safeguards into our compute infrastructure and embracing real-time automation to identify vulnerabilities more comprehensively. We are adopting machine learning technologies to identify anomalies in activity associated with security events and developing threats to ensure we have a secure and resilient software-defined network.

We consider machine learning in AI a form of data-powered learning systems. Traditionally, most analytics and policies for network workloads are deterministic, and moving to data-powered learning systems is a significant leap. We understand that leap at AT&T, which is why ECOMP was developed from the get-go to support both open-loop and closed-loop AI automation.

In open loop, we are able to capture network and cloud data and perform traditional rule-based and advanced ML analytics. Deterministic policies, written by humans, are then applied and tickets are issued when anomalies are detected. This is half automation and half manual intervention. As our operation team is becoming more confident with this process, they are starting to close the loop where actions are automatically managed by ECOMP while the platform continues to monitor and improve its actions. The same idea applies to security. The evolution from open to closed loop is vital to ensure systems are running optimally and as designed.

At AT&T, we are exploring numerous research opportunities and deployments for applying ML and AI for improving operational efficiencies, security of the platform, and data-powered policies, as well as enabling a new generation of cloud services and 5G that were not possible with legacy approaches.

TT: How does open source affect the security of automated systems? One school of thought says open source is inherently more secure because you’re not relying on a single vendor to find and fix vulnerabilities. Another says having the code in public view means hackers can peruse to their heart’s content.

MG: The answer depends on the open source. Having a code in open source with a large community involvement, such as ONAP, reduces the chance of intrusions because there is a significant team dedicated to reviewing, testing and examining the code around the clock. At AT&T, we have established frameworks and testing tools that scan every code ingested in our environment, ensuring we have intelligence in the platform that backs up to previous releases in case of any security intrusion.

Related posts:

• ONAP Poised to Transform Industry
• AT&T's Gilbert: The Goal of AI Is Automation
• AT&T's Gilbert Talks Virtualization
• AT&T's Gilbert: AI Is Very Real

— Tim Kridel, Contributing Writer, Telco Transformation