Gemini: A new frontier for the CRM Analytics Query Engine

It was four years ago that I wrote about a complete rewrite of the Query Engine powering dashboards and insights at the speed of light for all CRM Analytics customers. Every time a CRM Analytics customer loads a dashboard or explores a dataset, our in-memory engine, powered by our proprietary columnar store, retrieves the answer at sub-second speed. Every day, our engine runs tens of millions of these queries.

Well, at Salesforce, we make innovation one of our core foundational values, and we never stop. So, shortly after releasing the engine iteration codenamed Rocket, the team went immediately back to work on another generation of our engine, this time called Gemini.

I am happy to announce that after 2 years of work, the Gemini engine is now GA for all of our customers, delivering 25% faster response times at the P95 level across the customer base, with performance improvements of up to 60% for some of our largest and most challenging customers. Even more impressively, the number of timeouts our customers run into has been cut in half with the release of Gemini.

Note: When we talk about P95 we mean the speed at which 95% of the queries run – we keep the bar high for ourselves.

Trust is our number one value

As I mentioned earlier, at Salesforce, innovation is one of our core values. But the trust of our customers is our number one value. As we prepared for the deployment of the new engine, it was of the utmost importance for us to make sure not to disrupt our customers in any way. This means making sure that there would be no negative functional or performance impact on any of our customers.

So, how did we deliver innovation to our customers while making sure we don’t have a negative impact on the mission-critical aspects of their business that run daily on CRM Analytics? In other words, how did we replace a whole engine, while the engine itself was handling tens of millions of requests per day?

The answer is “Dark Launch!” With Dark Launch, we are able to execute real customer queries twice, the first time using the production engine to display the results to our customers, and the second time to test the new engine, creating a log that shows us if the results were the same and if the performance improved.

Note: As part of our commitment to trust, we never log actual customer data or query results, but just whether or not the results are identical between the engine the customer is using in production and the new experimental version.

This allows us to move forward with the release only when we have 100% certainty that there will be no negative customer impact, just better performance!

Learn more

To learn more about CRM Analytics visit our product page or connect with us on the Success Community. We constantly work on improving our engine and CRM Analytics, to deliver CRM insights at the speed of light, so make sure you connect with us and don’t miss on anything. The engineering team is also working on an upcoming blog that goes deeper into the engineering foundation of the new Gemini engine. As far as making sure you leverage this innovation, there is nothing you need to do – you already are on the new engine!

Written by

Antonio Scaramuzzino

See author's posts