Session

Flipping the Switch to Serverless: How McAfee Modernized Its Infrastructure and Halved Compute Costs

Overview

ExperienceIn Person
TrackData Warehousing
IndustryEnterprise Technology
TechnologiesDatabricks SQL, Unity Catalog
Skill LevelBeginner
DOWNLOAD SESSION SLIDES

At McAfee, we empowered 1,000+ end users and transformed our data platform — migrating over a third of production workloads to Databricks Serverless and unifying governance with Unity Catalog. Serverless compute put self-service analytics directly in users' hands, driving a 34% jump in monthly active users, a 30%+ lift in job efficiency, and $400K+ in projected annual savings — all delivered by a lean platform team. Our phased, data-driven migration playbook made adoption safe and reversible, using decision dashboards and clear revert-to-classic paths. To date, we've hit 65% infrastructure modernization — 90% of warehouses and 40% of jobs on Serverless.

In this session, we'll share how we governed workloads with Unity Catalog, selected high-impact migration candidates, overcame serverless limitations, and built the dashboards and tagging strategies that earned executive confidence. Attendees will leave with a proven blueprint to scale serverless adoption and unlock real value — increased velocity and reduced costs.

Session Speakers

Amisha Singh

/Delivery Solution Architect
Databricks

Mandar Pargunde

/Director of Data Engineering
McAfee

Full Summary

How McAfee cut Databricks costs in half by migrating 184 jobs to serverless

A real-world migration at McAfee shows how moving the right workloads to serverless compute can cut Databricks spend by roughly 50 percent and deliver dramatic runtime gains on select jobs, without pausing the business. The core message is operational: success depends more on governance, workload qualification, and guardrails than on the switch to serverless alone.

FAQ


Yes. Unity Catalog is required for serverless. McAfee entered with roughly 85 to 90 percent UC adoption. If workloads still rely on legacy Hive Metastore, complete that migration first.

Scala and R are not supported. Workloads depending on notebook-scoped JARs, RDDs, certain Spark configs, or caching behavior are hard blocks. Streaming jobs and anything using custom init scripts typically need refactoring.

Use performance-optimized for SLA-sensitive, low-latency workloads. Standard mode offers about 60 to 70 percent lower cost with a slightly longer startup, which fit McAfee's batch-heavy jobs. Match the mode to the workload.

Complete the security review, set up private connectivity if needed, implement tagging and budget policies, and capture cost and runtime baselines. Document a rollback procedure and treat it as a core safety feature, not a formality.

No. The reported 83 percent reduction reflects high-impact jobs, not the average. Some jobs may initially cost more until inefficient SQL is tuned. Serverless can expose defects and inefficiencies that classic clusters previously masked.