Yesterday, we organized a virtual panel on the role of ML and AI on Observability. Our guests for the panel include:
- Andy Mann from Splunk
- Aneel Lakhani from Honeycomb
- Eric Wright from Turbonomic
- Rob Hirschfeld from RackN
Andi and Eric took a more progressive view on the role of ML and AI while Aneel took a more realistic view. On the other hand, Rob took a contrarian position which made the panel interesting.
Key Takeaways on ML/AI In Observability
Some of the takeaways from this panel are:
- There is a clear need for Machine Learning to process Observability data so that the “What?” question can be answered. The large volume of data exhausts and the potential to unravel unpredictable behaviors are drivers for Machine Learning in Observability
- The problem is the lack of training data to train the models. Clearly, only web scale companies have data that are large and diverse to help train the machine learning models. But there were disagreements on whether that could really help other organizations with varying stacks and varying needs. But the majority of the panelists seem to think that there is a strong use case to tackle large observability data to find patterns that cannot be discerned by human beings. Personally, I feel that it is critical to use machine learning because humans cannot fully comprehend as our IT stacks get more distributed with containers, serverless functions, edge computing and IoT
- At a minimum level, machine learning can make the Observability data more useful for humans
- At present, the use of machine learning for Observability is at very early stages and it is going to take a long time before it becomes the norm. There are some organizations that are far advanced in tapping machine learning for Operations data but many are still not there yet
We are a strong believers in the role of ML and AI in Observability and in Operations. This virtual panel is a reality check on the market evolution. You can watch the entire conversation at Rishidot.TV.