beckmarkwith - Fotolia
BOSTON -- Seconds and minutes no longer qualify as real time when responding to network events. Instead, real-time response takes five to 10 milliseconds -- the time it takes to blink an eye.
That's how long an IT system should take to effectively respond to an event and make the necessary adjustments, said Rohit Mehra, vice president of network infrastructure research at IDC, during IDC's annual Directions conference this week. Although this response time sounds unachievable, it's not. The essential piece of the puzzle is network automation with AI integration, Mehra said.
Network automation is nothing new, evident in command-line interfaces, scripting, network fabrics, software-defined networking, software-defined WAN and, more recently, intent-based networking. These are the building blocks toward an autonomous, self-driving network, Mehra said.
"We're at the phase where AI is starting to enable a different degree of network automation," he said.
In traditional networks, network teams would configure the network to send alerts if an attack or change occurred. But these configurations didn't guarantee the team would respond to a possible alert immediately -- like if a network admin took a coffee or bathroom break, for example.
Additionally, traditional monitoring capabilities didn't always provide the visibility enterprises needed to meet application requirements. The network might not capture events or changes in real time, application context was missing, analytics were spotty and management tools were complex, he said.
Rohit MehraVP of network infrastructure, IDC
But with automation built into network infrastructure, networks can make real-time decisions and changes for application, network or security events. This real-time response becomes even more critical as autonomous cars, IoT and cloud networking grow more common. Further, third-party tools can integrate with this AI-enabled network automation to improve visibility and analytics.
"This gives us simpler, declarative management and verification policies, and it also gives us networks that can actually enforce and apply intent," Mehra said.
AI-enabled network automation use cases
Until recently, the industry lacked the computing resources necessary to support AI and machine learning tools. But as computing power and capacity increased, these tools gained the capacity to support automation needs with real-time response, Mehra said.
As these tools developed, Mehra said three use cases have emerged for AI integration with network automation:
- capacity planning and optimization;
- network operations; and
- visibility and security analytics.
Capacity planning and automation focuses on Day One of deployment. With AI-enabled network automation, IT teams get a real-world view of the network before deployment, gain improved network monitoring and provisioning, and they can dynamically optimize traffic flows, Mehra said.
In the second use case, network operations highlight the self-healing aspect of automation capabilities, along with real-time response, he said. These capabilities help the network better meet quality-of-service levels and application performance -- and they help the network adjust automatically.
Security is the final driving use case for AI and network automation. AI coupled with the network can better identify anomalies and pinpoint changes in network behavior by using traffic analytics and behavior modeling, Mehra said.
"You can benchmark what your normal behavior is, and the moment your analytics and visibility tools inform you about an anomaly, your system remediates right away," Mehra said. It filters down to the root cause of the issue.
Approach network automation pragmatically
Although AI with network automation is ripe with potential, Mehra recommended enterprises take a pragmatic approach to automation.
"Be judicious where you use automation," he said. "Find the right balance between human capabilities and automation capabilities. You need to get automation right when you're thinking about the network and using automation for mission-critical application needs."
This encompasses using clean, relevant and secure data when building AI algorithms, as flawed data and algorithms can be detrimental for enterprises. When implemented correctly, network automation with AI will augment IT capabilities, he added.