The network operation type tells us that AI will manage the network in the future. They also tell me that their vendor told them exactly the same thing. The good news is that it’s a kind of truth. The bad news is the same. The emphasis is on the modifier “sort-of”. To get the most out of AI network management, you need to get out of that hazy “kind of” zone, thinking about ants and farmers.
Ants can interconnect tunnels and levels of all kinds to build wonderfully complex anthills. Do worker ants have strong engineer ants to direct this process? No. Each of them simply performs their own simple tasks, and their instincts program them. There are actually ant engineers, but it is their own DNA that organized their work to achieve their goals. This is a bit like the behavior of most network AIs.
Each network consists of a series of technology “collections” like anthills. There is a collection based on vendor, device type, physical location, and connectivity. Looking at today’s network AI, it works primarily in collections. Maybe it manages edge elements like Wi-Fi or maybe SD-WAN or SASE. AI applications that manage collections have management goals built into DNA, design. If we are a Wi-Fi vendor, we know how Wi-Fi works and we are incorporating that knowledge into AI management.
Challenges arise when you stop thinking of collections as independent elements and start thinking of networks as collections of collections. The network is not an anthill, but an entire ecosystem with anthills inside, including many things such as trees and cows. Trees know how to become trees, and cows understand the essence of cowness, but what does they understand the ecosystem? The farm is a farm, not any combination of trees, cows and anthills. It is the farmer who knows what the farm is, not the elements of the farm or the suppliers of those elements. In your network, dear network operation type, that farmer is you.
In the early days, AI developers clearly allowed the separation between the knowledge engineers who built the AI framework and the experts in the areas where knowledge formed the framework. In software, especially DevOps, management tools aim to reach the target state. This shows where cows, trees and ants fit in the farm analogy. If the current state is not the target state, they do something or move. Around to converge on the goal. This is a great concept, but for it to work, you need to know what your goals are. At the enterprise network level, it requires knowledge that Wi-Fi professionals have subliminally introduced into Wi-Fi AI management tools. If AI vendors don’t know how that knowledge is acquired, their AI won’t help.
Keep in mind before deciding that your hopes for AI will be shattered forever! Many network operation types are completely satisfied with the AI that manages the collection of technologies that make up the network. After all, why worry about coordinating Wi-Fi and SD-WAN management when what happened on one side can’t be improved by shaking the other? This Collection-If your AI model fits your needs, you don’t have to be at home.
A good way to see if an ant is okay (at least for network AI) is to ask if the tech collection is really atomic (fully independent and self-contained). That is. It comes down to the visibility and control of your AI. Basically, collection-specific AI maintains itself. Ideally, AI collection ants should do their own thing without stepping into each other’s activities. It is not desirable for an AI in one place to look into another collection and react to the condition, or for two AI collection processes to tackle the same problem at the same time without coordination.
If the solution to a problem in one collection could involve something to another collection, you need to launch AI to cover that combination. So if you’re looking at an expensive and overworked network operations center that manages ecosystem issues and wondering if AI can break coffee for everyone, you need deeper insight into the vendor’s AI claims. .. This isn’t easy for businesses, as more than three-quarters of the people I talk to this year say they don’t have much in-house AI expertise. Many feel at the mercy of vendors who promise great things and don’t seem to provide enough of what they expect. Is there anything a company can do?
The easiest way to use AI throughout the network ecosystem is to look for strategies like the old “manager-manager” approach. In modern terminology, it can be called this intent modeling. If each technology collection can be treated as a black box that models its behavior against an SLA and its AI process functions to enforce that SLA, then each of those collection AI tools needs to fail. Just generate a report. For higher level packages. The package allows you to decide what to do if you have problems that go beyond a single collection of technologies, or if one collection just throws a towel and you need to consider a higher level of fix.
The challenge here is to find out what the goal is and how to return to it when a problem occurs. Do you remember the experts and knowledge engineers in those areas? It’s difficult to build an AI solution for a network because all networks are a little different and only users know what they consider to be “good” or “bad.” Some AI tools may provide machine learning (ML) capabilities that allow NOC personnel to look over their shoulders and learn what to do. You may also want to use a baseline that network vendors know to represent the usual options and common remedies.
Both approaches have some problems. Machine learning can be time consuming and can further waste NOC resources while the AI system is learning its mission. Vendor baselines work best when the majority of your network consists of equipment from one vendor. Both can be adjusted, but both can violate the behavior of adaptive networks.
IP networks basically use topology discovery and do their own thing. Even NOCs are difficult to influence routing. They often have to plan new MPLS routes to do traffic engineering, which is unlikely to be done by AI. Some companies (including Google) have adopted Software Defined Networking (SDN) to provide centralized control of routing, and AI can control the network by controlling the SDN controller.
AI in network operations goes back to the combination of events to notify changes and implement effective responses. At any level, future AI vendors should be able to discuss how their products collect information and implement insights. Dig into the details of these two things. Because whatever the magical AI claims to work, it won’t work without these two elements. Be a farmer, not an ant.
Copyright © 2021 IDG Communications, Inc.
Is your network AI as smart as you think?
Source link Is your network AI as smart as you think?