Improving Network Automation & Security With Artificial Intelligence
Communication service providers (CommSPs) are already saving money and generating revenue from network transformation investments. There is an expectation these benefits will continue to increase as NFV functions scale across the various elements of the infrastructure - enterprise, radio access network, wireless core, cable and cloud. New 5G and edge computing use cases promise to deliver new revenue along with even more data that must be moved, stored, processed and analyzed.
The industry is looking to Artificial Intelligence (AI) and Machine Learning (ML) to enable CommSPs to solve problems and unlock value for their own business operations and their customers. As an example, distributed AI based on reinforcement learning will play a key role in building automated and self-managed networks. Let’s explore a few examples where AI is applied in the network for improving operations including Closed Loop Automation, Encrypted Traffic Analytics, and Radio Optimization for 5G beamforming.
AI Supports Closed Loop Automation, Network Optimization and TCO Benefits
The growing complexity of network infrastructure, combined with the low latency and determinism associated with next generation services, makes it impossible to deploy and manage networks based on traditional network management methods and static policies. These methods do not scale and have a significant impact on the operational costs. Networks have to be dynamic and automated.
AI and ML advancements make closed loop automation possible with NFV, which will be essential to the remote monitoring and management of thousands of network edge locations and billions of connected devices. When integrated with an operational support system (OSS), AI-powered network monitoring and predictive network analytics capabilities will detect network anomalies and faults, analyze root causes and trigger fault recovery/failover before failures actually occur in the network. Self-monitoring, self-managing and self-healing networks make it possible to dynamically adjust resource allocation and power consumption, which results in the reduction of expensive technician deployments.
At Intel, we are focused on three key elements for AI-enabled closed loop automation – 1) platform telemetry 2) trained and curated AI models and 3) orchestration to drive real-time inferencing. In fact, Intel recently demonstrated how traffic prediction models can be used to dynamically control the power states of the CPUs and platform components to reduce the power consumption – a major contributor to operational expenses.
Encrypted Traffic Analytics
Google estimates that over 90 percent of all Google web traffic is encrypted, and Gartner estimates that 80 percent of enterprise web traffic is encrypted in 2019. While that’s great progress, malicious actors may use encryption techniques in order to evade detection and disguise malicious attacks. Today most enterprises are unable to look into encrypted traffic and perform deep packet inspection for malicious content, but AI promises new safeguards against these attacks.
By populating AI models with telemetry data, such as sequence of packets, packet boundaries, nature of compute operations and memory access patterns, we can effectively and efficiently perform real time intrusion detection, network isolation and preventive actions on encrypted traffic. In fact, Intel is in a unique position to help companies harness much of this telemetry data from hardware platforms and use that data for the purpose of training AI models for security analytics.
Radio Optimization for 5G beamforming
AI innovation promises to enable wireless coverage and capacity optimizations, advanced traffic management, dynamic distribution of users across frequencies to improve user experience, dynamic radio resource management, beamforming configuration, multi-radio access traffic steering/management, service aware resource management for network slices, and much more.
One of the biggest innovations in wireless networks is MIMO technology (multiple input multiple output) which expands the capacity of a radio link using multiple transmission and receiving antennas. With this technology comes the challenges of managing the beam patterns and minimizing power consumption. A deep learning model trained with historical time-series data can continuously optimize beam patterns based on device type, user locations, traffic behavior, interference and other parameters. Power consumption is one area where CommSPs can apply this innovation for significant operational savings. Similar to the closed loop automation described earlier, network operators can apply trained, machine learning models, based on traffic patterns, to predict lower traffic conditions and automatically lower the frequency and power consumption (low power state) of the radio units.
Push the Limits of Infrastructure, Operations and Development with AI
AI and ML have enormous potential for improving network performance, network automation and security. We are working with partners and customers across a number of industry ecosystems to solve many network optimization and automation problems utilizing the power of AI. I invite you to join this dynamic community to push past the limits of traditional network infrastructure, operations and development to usher in the data-centric 5G era.
— Rajesh Gadiyar, Vice President and CTO, Network Platforms Group, Intel Corporation