QuantaStream

Bitmap-native analytical SQL.

QuantaStream explores how SQL can execute over compressed bitmap domains, BSI-backed values, and relationship vectors.

It is quietly taking shape as a systems project for analytical workloads, late materialization, and set-based query execution.

View the white paper Start a conversation

Why QuantaStream?

A MySQL-shaped front door for bitmap-native analytics.

The backdrop is simple: analytical SQL often spends too much time re-discovering sets of rows. QuantaStream stores those sets directly as compressed, queryable bitmap domains, then moves SQL execution toward set algebra, relationship vectors, and late materialization.

On the SF1 TPC-H read-only benchmark, a single-node QuantaStream server showed a clear advantage over a 16-vCPU MySQL reference. The strongest wins came from the query shapes that matter most for analytical systems: grouped scans, graph-style joins, and relationship-heavy predicates.

The point is not to clone MySQL internals. The point is to offer a familiar MySQL-compatible endpoint while changing the physical representation underneath it.

1

Streaming-ready HTAP shape

QuantaStream is being built so operational ingestion and analytical queries can share the same compressed representation.

2

Domain-specific data mappers

Custom mappers can encode timestamps, enums, scaled numeric values, identities, and application-specific facts into directly queryable compressed form. For example: A UUID is rendered as a 36 byte string. Internally it can be a 16 byte long integer.

3

Relationship vectors as first-class data

Instead of treating joins as repeated row reconstruction, QuantaStream can preserve relationship structure and push work down into bitmap operations.

4

A path from one node to distributed execution

The same engine direction supports a single-node MySQL-compatible server and a distributed mode for larger analytical workloads.

Single-node QuantaStream vs MySQL

SF1 TPC-H read-only benchmark, 16 vCPU AWS hosts. Lower runtime is better.

Bitmap-native advantage
Q5 graph join 6.4x
Q1 grouped scan 8.0x
Q3 revenue 2.8x
Q21 sibling exists 3.1x
Q6 filter sum 2.9x
MySQL QuantaStream

Compatibility coverage is still expanding, but the endpoint shape is intentional: keep the familiar SQL surface while replacing row-first execution with compressed-domain execution.

Distributed benchmark

So how does QuantaStream scale?

The latest three-node AWS run moved the data nodes to r7i.4xlarge and pushed the loader to 12 workers. At 3x the SF1 data, the read-only suite completed in 2.4x the time and the broader profile suite came in at 2.6x.

Load elapsed

Full TPC-H distributed load after tuning loader concurrency.

SF1 7.0m
SF2 13.0m
SF3 21.7m

3.11x SF1 load time at 3x data, with the memory boundary cleared.

Read-only suite

The compact correctness/performance suite.

SF1 8.0s
SF2 14.2s
SF3 19.6s

2.44x SF1 time at 3x data.

Profile suite

Relationship-heavy paths with richer query coverage.

SF1 16.2s
SF2 28.4s
SF3 42.2s

2.61x SF1 time at 3x data.

The useful part: query scale

SF3 divided by SF1 runtime. The dashed marker is the 3x linear target; lower is better.

3 data nodes · SF1/SF2/SF3 · final tuned runs

Profile q3 2.82x
Profile q5 combined 2.73x
Profile q5 formal 2.72x
Profile q5 same 2.57x
Readonly suite 2.44x
Readonly q19 1.93x
Profile q19 1.56x

About the founder

Guy Molinari

Principal Engineer and Founder
Playa Zancudo, Costa Rica

Guy Molinari

I spent 21 years at The Walt Disney Company as a software engineer, architect, and engineering leader. My early work included Disney's Tea language compiler and Barista servlet container, helping migrate Disney web applications to Apache Tomcat, and leading development of FastCast - an asynchronous server cluster designed to support approximately 20 million concurrent connections for ESPN live-score feeds and GameCast sports visualizations.

In 2019, I moved to Disney's Data Platforms organization and began developing bitmap-based technology for managing and querying advertising revenue and inventory data. I am the sole inventor of U.S. Patent 12,086,136, Techniques for Executing Join Operations Using Bitmap Indices, and a named inventor on multiple other U.S. patents covering bitmap-oriented data management, adaptive caching, and related technologies.

In 2021, I became Senior Manager of Data Platforms Engineering at Disney Streaming, where I led a team of 13 engineers developing the Data Activation Framework.

I retired from Disney in 2025, but not from engineering. I founded QuantaStream to continue exploring an idea that began during my time at Disney: that bitmap and bit-sliced indexing can provide a fundamentally different foundation for distributed SQL, real-time ingestion, and analytical query processing.