How Palo Alto Networks uses BigQuery ML to automate resource classification:

Palo Alto Networks firewall is a leading cybersecurity company transforming the cloud-centric future worldwide. Our products are trendy and world-renowned, with over 170,000 projects available on Google Cloud. Our excellence team handles all our central cloud operations. Our Palo Alto Networks world has been growing significantly ever since. With such expansion, we aim to ensure the process stays cost-effective while maintaining proper cloud hygiene and maximum security. However, the most challenging part of our journey was when we had to identify and match each project with its suitable team, cost center, and environment, which was a significant starting point for our work. This was when we took the help of an automated labeling system to cover almost 95% of the tasks regarding tagging for the team, owner, cost center, and environment. But completing the last 5% was even more challenging. Finally, we decided to use machine learning to make our work more effective and the operations more seamless. This is it. This is the backstory of how we embarked on our journey of using BigQuery ML for automatic resource classification.

Prototyping projects took only two hours with BigQuery ML:

It was extremely difficult for us to identify the owner, cost center, or environment for all our cloud projects, which covered a large number. And without proper identification we could not examine exactly how much money our team of expert engineers were spending on each of the cloud based firewall resources.

This whole process used to take weeks when the finance team had to go through hundreds of projects one by one and contact the owners to accurately assign team owners on the dashboards and reports. With such immense pressure, we decided to only sort out the high-cost projects and leave the low-spend projects unidentified and without any ownership label.

But whenever questions about any project ownership arose, our team used to look for keywords in an attempt to identify which team was linked to the said project. This method often required us to trust our intuition based on keywords, and we ultimately realized that it was high time to take advantage of machine learning to automate this process.

Google Cloud advised us to give BigQuery ML a try for prototyping the project which only took a few hours to complete, that too with 99.9% accuracy. Then we further moved on to test it on our other projects and got correct label results each time.

BigQuery ML can identify projects with an accuracy rate of 99.9%

With BigQuery ML, we can now identify which project belongs to which team, cost center, or environment with a precision of 99.9%. After learning the efficiency and power of

BigQuery ML, we are planning to extend its benefits to other teams and use cases further. One such example is implementing this model in the domain of finance and information security, where our teams require additional information about the projects. Moreover, if we find any suspicious online breach activity in one of our unmapped projects, we can quickly identify the owner of the affected project by leveraging this model.