.shock - Fotolia
Self-driving cars and computer networks may not have a lot in common, but that could change if Juniper Networks gets its way.
The supplier of network security products has been touting the concept of self-driving networks that configure automatically according to needs of an enterprise, as well as fend off cyber attacks.
In an exclusive interview with Computer Weekly, Kireeti Kompella, chief technology officer and senior vice-president of Juniper Networks, said self-driving networks are a direct response to the growing cost and complexity of managing corporate networks.
“The cost of managing networks, which are getting bigger, is growing,” Kompella said, adding that it has also become harder to find networking experts. “How can companies become agile and roll out new services quickly, if operational expenses are going up and the talent pool is shrinking?”
While automation may be the answer to these woes, Kompella said much of network automation today is akin to putting automatic transmission into a car. Instead, he called for the development of networks that run on their own, like self-driving cars do.
In a research note, Cliff Grossner, senior research director at market research firm IHS, said ML – likely leveraging neural networks – can be used to train AI systems to recognise patterns in vast amounts of real-time network telemetry data.
With early recognition of developing network performance issues before they occur, policy-driven AI systems can then take automatic action to change the course of events, avoiding outages that affect the user experience, he said.
While network performance and reliability are likely to improve with self-driving networks, Kompella said the killer application is in cyber security.
Read more about network automation
- Detailed network micro-segmentation using SDN and NFV gives service providers opportunities to create and provide more flexible and effective security services to their customers.
- SDN and DevOps may both have a role in automated network device management, but compared to SDN, DevOps has a completely different mission, our expert explains.
- You might not always see it, but that doesn’t mean network automation isn’t there... and as its footprint grows, the command line interface could disappear.
- The Open Networking Foundation unveils its Open Innovation Pipeline to guide the next phase of development of software-defined and virtualised networking tech.
“A network failure is not necessarily urgent, but a security breach is,” he said. “Having AI and machine learning on your side, reducing false positives and finding the root cause of an attack and knowing how you can block it, become really important.”
The key to a self-driving network is data that reveals the behavioural norms of users, applications and devices across different network layers.
“Whether it’s a core router, an edge router or an IoT [internet of things] device, we need data to know what’s considered normal operations,” he said, adding that this will let organisations such as telcos and cloud service providers run their infrastructure at nearly full capacity, rather than make provisions to handle unexpected surges in demand.
Kompella said behavioural data can be generated through streaming telemetry, where bots are programmed to stream – rather than pull – data about network flows, applications and devices in real-time. “It’s very lightweight, and the fidelity of the data is higher,” he said.
Streaming telemetry is currently available for Juniper’s edge routers, though the company will be extending it to applications that sit on corporate servers through Contrail, its software-defined networking platform.
Grossner noted that while Juniper is seen as a thought leader in using AI and ML to provide high performance and secure networks, it is still early days in delivering on its vision.
“Much development needs to be done on understanding the nuances of embedding AI and ML into networking. Juniper’s Contrail will provide an excellent perch point to introduce AI and ML creating a self-driving network.
“The challenge for Juniper will be maintaining the sustained AI and ML investment required until we see results that can be monetised,” he said. ...