In part 3, you realized the following:
- How many experts are needed to identify the following?
- Solution architecture/topology
- The resources required for each transaction
- The key performance indicators for each resource
- The key log messages for each resource
- The threshold for each key performance indicator
- Maps between transactions and resources
- Relationships and dependencies between resources
- You also realized how much time and effort the above activities require.
The time and effort become more prominent as new issues come up, new key performance indicators are identified, new thresholds are set, transaction to resource maps are updated, and relationships and dependencies are redefined. This is because the process of performing the activities above is iterative. The idea is that the solution will initially need more baby-sitting in production until its wrinkles are ironed out.
What happens if we infuse AI into the identify activities above? Clearly, infusing AI into these activities is easier said than done.
The main component of the AI solution will make use of the following systems that implement a few of the identify activities above and that may already exist in the traditional (non-AI) operations environment:
- Solution architecture/topology software: This software is used to provide the topology to the main AI component.
- Monitoring Software: This software tracks the key performance indicators, corresponding thresholds and fires alerts when thresholds are reached. The main AI component is one of the consumers of these alerts.
- Log streaming software: Software that feeds the log messages to the AI component.
Although there are other inputs to the main AI component, I am presenting the need to infuse AI in the identify activity here. The need to infuse AI in the other activities will be presented in future posts.
Here is a relatively simple example of how AI infusion can help identify new KPIs or new messages. The main AI component (machine learning model) gets trained on a normal state of the solution long enough to know what the normal state is. This training includes reading all inputs from the above systems. Once the AI model is trained, it watches all incoming data from the above systems. The AI model will be able to identify what is new (something that it has not seen before). Consequently, it flags it to the experts to look into. After looking into it, the experts may decide that they need to create a new KPI or a new message to track. Clearly, without the AI fusion into the identify activity, the identification of such a new KPI or a new message can be very time-consuming as the amount of data to visually inspect can be very overwhelming in a typical solution.
In the next blog posts, we will continue exploring the need to infuse AI into IT operations.#Featured-area-1#Featured-area-1-home#AI#ArtificialIntelligence#itops#watsonAIOps