There are many enterprises in the database monitoring screen, monitoring screen this thing in the development of China's information system in the past 20 to 30 years has been constantly evolving. The large screen is becoming more and more beautiful, and the data above is becoming more and more valuable, but I think the development of monitoring large screens has reached a bottleneck point. Because with the faster and faster development of information technology, the construction of information systems is becoming more and more complex, both the number of information systems and the complexity of indicators have exceeded the limit of visual analysis. Now the monitoring screen has become a form to a large extent, the content of the display, the purpose of the display, how to do monitoring through the large screen, has been unable to really play a role.

A while ago, Younuo's Ao cold total passing Nanjing, because time is very limited, can only do a small gathering in the coffee shop under the stone city. They are the first batch of enterprises to do digital twins in China, and I also mentioned to him in the early days that the colorful animation seemed not so intuitive for operation and maintenance. If you really want to let the operation and maintenance personnel understand the system and understand the status quo of the system, the digital table may be the most intuitive and effective. However, at that time, customers liked realistic simulation display, and they were the best products in domestic simulation display.

When we talked about it again, he showed me some of their latest dynamic vision technology. I found that the fancy 3D interface was gone, replaced by a very vivid animated box that made it easy and intuitive to see how the business was performing. Combining the algorithm and model, the simulation on the big screen is no longer emphasized, but the real internal logic of the business is simulated more deeply, and the monitoring personnel can intuitively feel the status of the business operation and find the possible blocking points and risks of the business.

Another big change is that relying on the rich data collected in the system, through the business simulation playback function provided by the Yoplait dynamic vision engine, this system can also play a huge role in business problem analysis and business fault tracing.

My first impression was that this was what a digital twin should look like. Digital twin systems are built to solve problems, not just to satisfy people's visual senses. More importantly, the system can help us to analyze the business, find and solve problems in the business.

Back to the field of database operation and maintenance, most of the database monitoring systems we currently build are based on traditional network management thinking, and the main purpose of collecting indicators is to show. The data collected needs to be made into a dashboard before it can be used. So now when it comes to database observability, a lot of people talk to me about Grafana. And in fact, that's not at all what I know about observability.

There are three major problems with Grafana-based observability: The first is that as more and more database systems become available, if a company has hundreds or even tens of thousands of databases, who do you want to show all those dashboards to? If you have a friend to talk to, Grafana can set the alarm rules, there is an alarm need to look at the dashboard. So the second question is, how can Grafana be more accurate? Do you rely on baseline thresholds? Will thousands of false alarms a day break you? If you have solved the second problem, then the third problem comes again, in the era of database localization, do you have a DBA who truly understands these databases? I'm afraid that few of the original after-sales service personnel are real experts in their own database products. Even in the era of operating and maintaining Oracle, front-line monitoring personnel can not be experts, experts are generally in the third line.

I think that traditional monitoring has entered a dead end, and in fact, AI monitoring is the future of monitoring systems. In fact, both generative AI and traditional AI algorithms can mimic the way experts think to interpret and analyze data. The collected monitoring data does not necessarily need to be displayed using the dashboard, but it must be analyzed by AI algorithms. The problems found after the small model calculation are then handed over to the generative AI based on the large language model for summary analysis, and finally the problems and findings are summarized and related alarms are generated, which may be the basic construction idea of the future monitoring system.

The monitoring system that invests a lot of money to build will eventually play its due role. One minute to find the problem, five minutes to locate the problem, 10 minutes to solve the problem, this is the current bank IT operation assurance goal, probably also the vast majority of critical business system operation and maintenance goals. To achieve such a goal, it is almost impossible to rely on traditional human flesh operation and maintenance, human flesh monitoring. Only when the person sitting behind the monitoring screen becomes an AI agent, this goal can be truly realized.