Moloco Apache Pinot Priority Case

Moloco — DSP: 10 Minutes → Milliseconds

Moloco cut query latency from 10 minutes to milliseconds by migrating its $1B+ ad platform to Apache Pinot + StarTree, now serving 6M queries/second.

Architecture diagram: Moloco — DSP: 10 Minutes → Milliseconds

Scale

$1B+ annual ad spend, ~10PB data/day, 6 million queries/second

Before

10-minute delays in performance analytics portal due to high cardinality and complex queries

After

Query latency from 10 minutes → milliseconds with Apache Pinot + StarTree managed service

Key Insight

6M QPS at millisecond latency is the upper end of what any system handles — this is the reference for 'what does Pinot do that nothing else does.'

In a Snowflake Conversation

6M QPS at millisecond latency is the upper end of what any system handles — this is the reference for 'what does Pinot do that nothing else does.'

My Read

Practitioner commentary coming soon.

Apache Pinot StarTree DSP query latency high QPS

Relevant Conversations

Streaming OLAP Why Not Snowflake