LiquidM — DSP with 50+ Targeting Dimensions
LiquidM runs a DSP handling millions of requests/second, filtering on 50+ targeting dimensions simultaneously within 100ms using Apache Druid.
Scale
Millions of requests per second with <100ms response requirement
Before
Systems unable to handle 50+ dimension filtering at millisecond latency
After
Apache Druid filtering on 50+ dimensions simultaneously — example: 'All male iPhone users over 40 within one mile of a Starbucks'
Key Insight
Multi-dimensional filtering at millisecond latency on high-cardinality data is Druid's core strength — Snowflake can't do this interactively at this latency.
In a Snowflake Conversation
Multi-dimensional filtering at millisecond latency on high-cardinality data is Druid's core strength — Snowflake can't do this interactively. This is the use case where Druid wins unambiguously.
My Read
Practitioner commentary coming soon.
Relevant Conversations