Our goal is to use machine learning – data, feedback and prediction models – to reduce MTTR. This has been a multi-year journey for Workspot and we are very excited to share our journey.
There are 3 keys to AI/ML for DEX:
- Comprehensive Cross-customer Data
- Community Feedback
- Prediction Model
Let’s put this in the context of what IT wants to accomplish with DEX: resolve problems faster or reduce MTTR (Mean Time To Resolution). At a high level, the workflow for problem resolution starts with Detection, then Root Cause Analysis, and finally Resolution.
Most companies have 3rd party tools for problem detection for VDI, e.g., ControlUp, Lakeside Software, and others. These tools capture data that assists in root cause analysis, often performed manually by multiple functional teams in servers, storage, networking, VDI, desktop operating system, etc.
To reduce the time to resolve a problem, we need to incorporate machine learning into the problem resolution workflow so that when we detect a problem, we are able to provide a predicted root cause and resolution.
Let’s discuss the 3 keys in more detail:
- Comprehensive Cross-Customer Data: The need for comprehensive high-quality data is well understood – you need to have data about the client, network, gateway, control plane, desktop operating system, etc. Cross-customer is interesting in the context of machine learning, because you can learn from problems seen at other customers to predict root cause for your own problems. For e.g., if a bad Windows update is causing crashes, or a region of the cloud is down, or an NVidia driver update is causing application behavior issues, you can learn from other customers’ problems to assist in root cause analysis and resolution.
- Community Feedback: Let’s say a customer or Workspot support is able to resolve a problem seen by one of their users. We want them to train the machine learning model by providing that feedback directly to the system, so the model becomes better for everyone. We don’t anticipate that every customer will provide such feedback, but as the amount of data and feedback grows over time, the quality of the predictions will continue to improve.
- Prediction Model: Data + RCA/Resolution Feedback is used to create a prediction model for future problems. And more data + more feedback will result in better models.
We will share some examples from the implementation in future posts.
