Michelin cuts project costs by 30% with MongoDB

THE CHALLENGE

Scaling the monitoring of global flows

Many will recognize it as a producer of tires, and while this is a core part of its business, there is much more to Michelin.

“We don't just make tires. We have 130,000 Michelin employees across a range of industrial and high-technology sites,” explained Aude Malabat, Product Owner and Database Team Leader at Michelin. “We’re interested in many other things, the Michelin Guide for restaurants in particular.”

For an operation of Michelin’s scale, keeping track of internal operations is essential. To ensure smooth operations, it relies on an End-to-End (E2E) application that monitors internal data flows globally, historically tracking messages across IBM MQ and Kafka. This business-critical application allows DevOps teams to replay integrations and track error statuses. However, the application’s legacy on-premises infrastructure was causing severe bottlenecks. 

Michelin previously utilized a custom MongoDB configuration spread across heavy servers, each with 8 CPUs and 314 GB of RAM. Managing version updates, server maintenance, and complex cluster configurations ate into valuable team resources and complicated daily operations. And even with this massive infrastructure footprint, the system still struggled during peak times. 

“We always get twice as many messages at the end of the month,” explained Yannick Raffin, Transversal Technical Leader at Michelin. “Sometimes we would even crash the on-premises cluster because the data volume was just too great.”

To complicate matters further, Michelin couldn’t simply take the system offline to carry out the upgrade process.

“We had a migration plan, but we also had a constraint, which is that the system can’t be shut down,” said Raffin. “Applications have to keep running and we have our end-of-month billing flows. Losing those would create even greater issues.”

OUR SOLUTION

A scalable, managed solution in the cloud

Michelin saw that the best way to achieve the elasticity and simplicity it needed was to move to the cloud. The logical decision was, therefore, to migrate its E2E application to MongoDB Atlas. 

Working closely with experts at both MongoDB and the MongoDB Cloud Migration Factory, Michelin's database team orchestrated a zero-downtime migration strategy using mirroring and a dual-run approach to ensure that data flowed seamlessly to both the legacy system and to MongoDB Atlas.

“With the mirroring, you reach a point where it retrieves the data simultaneously from the database while the database itself is powered,” said Raffin. “It then sends the data to MongoDB Atlas, which recreates the indexes and data. When the lag in that process reaches zero, the mirroring is perfect and we can turn everything else off.”

Moving to Atlas immediately solved the team's cluster management burdens, as version and server updates were now fully managed. To tackle the disruptive end-of-month data spikes, Michelin enabled Atlas Autoscaling on an M60 NVMe cluster. 

This allowed the database to automatically scale compute and storage resources to handle the doubled message volume during billing periods, and seamlessly scale back down once the overload passed, completely eliminating the on-premises bottlenecks.

Michelin also leveraged Atlas Search to replace an old, highly customized search mechanism that was previously developed, as Raffin put it, “the old-fashioned way”. 

By implementing MongoDB Atlas Search, Michelin was able to streamline how the E2E application indexes and queries messages, completely removing the need for resource-heavy custom lists that were previously inflating the database size and causing severe index bloat.

“Our aim was to optimize this as much as possible,” said Raffin. “The monitoring on Atlas Search really helped out here, alerting us to slow queries and where to create new indexes to resolve them.”  

OUTCOME

Smaller messages, reduced costs

The migration to MongoDB Atlas has transformed Michelin's flow monitoring capabilities, delivering major improvements in performance as well as significant cost savings.

Implementing Atlas Search enabled Michelin to drastically optimize its data model and streamline its query structures.

“Previously, the application relied on 26 separate indexes to function. In the end, we got that down to just six,” explained Raffin. “MongoDB has enabled a 70% reduction in the size of messages we send. That creates significant performance gains in both CPU and disk space.”

Even more significantly, Atlas Search has also allowed Michelin to achieve a 30% saving on the overall project cost. And beyond the impressive performance metrics, the move to a fully managed cloud database has unburdened Michelin’s technical teams, shifting their focus away from manual cluster management to higher-value tasks.

Michelin is now well positioned to scale its vital E2E monitoring application effortlessly alongside its continuous global industrial growth.

“MongoDB Atlas gives us peace of mind because we’re working with a full SaaS solution,” explained Malabat. “It also means we can be sure that we’re working at the optimum cost. It’s an important win.”