Trino: The Engine for the Data Lakehouse

1.0 Executive Summary

Trino, formerly known as PrestoSQL, is a high-performance, distributed SQL query engine for big data analytics. The core value proposition of this technology is its ability to run high-performance, interactive SQL queries on data “in-place,” directly at its source. This capability allows organizations to query data across disparate systems—such as data lakes, relational databases, NoSQL systems, and cloud or on-premises environments—without performing costly, complex, and slow data movement (ETL).¹

Unlike traditional data warehouses that require data to be co-located in a proprietary system, Trino operates on a “query-in-place” principle. It provides a single point of access and a unified ANSI SQL interface to query data where it lives, making it a foundational component of modern data architectures like the data lakehouse and the data mesh. As a query engine, it is not a database and does not store data itself; rather, it is a sophisticated execution engine that can coordinate and execute queries (OLAP)² across one or more heterogeneous data sources.

This report provides a comprehensive technical overview of Trino, including its history, core architecture, and its critical role as a federated query engine. It analyzes its relationship with key ecosystem technologies, including table formats (Iceberg, Delta Lake, Hudi) and competing platforms (Databricks, Athena, BigQuery). The central conclusion is that Trino’s open-source, flexible, and connector-based architecture makes it a powerful and strategic tool for unlocking the value of distributed data and avoiding vendor lock-in.

2.0 History and Origins: The Presto-Trino Split

2.1 Presto at Facebook

The project’s origins trace back to Facebook (now Meta) in 2012.³ Martin Traverso, Dain Sundstrom, David Phillips, and Eric Hwang began development on a new distributed SQL query engine designed to run interactive analytic queries against the company’s massive 300-petabyte Apache Hadoop data warehouse. This project, named Presto, was open-sourced by Facebook in 2013.

2.2 The Fork: PrestoSQL (Trino) vs. PrestoDB

In 2018, the original creators (Martin, Dain, and David) left Facebook to found a new company, Starburst Data, built around the open-source project.⁴ This led to a fork in the development:

PrestoSQL: The original creators continued development on the original codebase under the name “PrestoSQL,” focusing on community-driven, open development.
PrestoDB: Facebook continued its own internal development on the original “Presto” (now PrestoDB) codebase, with contributions from other large tech companies like Uber and Twitter.

This split created significant confusion in the market. To resolve this, the PrestoSQL community announced in December 2020 that it was rebranding the “PrestoSQL” project to Trino.⁴ The Trino Software Foundation was established to manage the project, with the original creators at its head.⁵

Today, Trino is widely considered the direct successor to the original Presto open-source vision, with a more active and open community.⁶ PrestoDB development continues, but it is more closely tied to the specific needs of Meta and its partners.⁸ For the remainder of this report, “Trino” will be used to refer to the project formerly known as PrestoSQL.⁷

3.0 Core Architecture

Trino is a massively parallel processing (MPP) query engine. It follows a classic coordinator-worker architecture, similar to other MPP systems designed for Big Data.⁹ The system is composed of two main types of nodes: a Coordinator and multiple Workers.¹⁰

3.1 Coordinator

The Coordinator is the “brain” of the Trino cluster. It is a single server responsible for:

- Parsing & Analyzing SQL: A client, such as a BI tool or JDBC/ODBC driver, sends an ANSI SQL query to the Coordinator.
- Query Planning & Optimization: The Coordinator creates a logical query plan, optimizes it using a Cost-Based Optimizer (CBO)¹¹, and translates it into a distributed execution plan. This plan consists of multiple stages to be executed in parallel across the workers.
- Task Scheduling: The Coordinator assigns tasks from the execution plan to individual worker nodes based on data locality and resource availability.

* It does not participate in the actual data processing.¹¹

3.2 Workers

Worker nodes are the “muscle” of the cluster. They are responsible for executing the tasks assigned by the Coordinator.

Data Retrieval: Workers fetch data from the source systems (e.g., S3, MySQL, Kafka) via Trino’s Connector mechanism.
In-Memory Processing: Trino is an in-memory, pipelined execution engine. Data is processed in memory and streamed between stages (from one set of workers to another) without being written to disk.¹² This in-memory, pipelined approach is what gives Trino its high speed, but it also means it is resource-intensive, requiring sufficient RAM.
Data Shuffling: Workers exchange intermediate data with each other as required for operations like joins and aggregations.
Returning Results: The final stage of workers returns the result set to the Coordinator, which then streams it back to the client.¹²

3.3 The Connector Architecture: Data Federation

The most important concept in Trino’s architecture is data federation. Trino itself does not know how to talk to S3, MySQL, or any other data source. It achieves this by using a pluggable Connector interface.¹⁰

A Connector is a “driver” or “adapter” that teaches Trino how to communicate with a specific data source.¹³ It is responsible for translating the Coordinator’s plan into the native language of the source system (e.g., generating a SQL pushdown for a relational database) and fetching the data.¹⁷

This architecture is what enables Trino’s core use case: cross-source federation. A user can write a single SQL query that joins a Hive table in a data lake, a PostgreSQL table in a transactional database, and a Kafka topic from a streaming platform. The Coordinator will plan the query, and the Workers will use three different connectors to fetch data from all three sources, performing the join in memory.

4.0 Trino’s Role in the Modern Data Stack

4.1 Data Lakehouse and Table Formats (Iceberg, Delta, Hudi)

Trino is a key component of the open data lakehouse.⁵⁷ A data lakehouse combines the low-cost, open storage of a data lake with the transactional, ACID-compliant features of a data warehouse. This is achieved through open-source table formats like Apache Iceberg, Delta Lake, and Apache Hudi.⁵⁸

Trino is not a table format. It is a query engine that can read from and write to these table formats using its connectors.^{54,⁵⁹,⁶⁰ Trino’s support for all three major open formats makes it a “universal” query engine for the lakehouse, allowing organizations to avoid being locked into a single vendor’s ecosystem (e.g., Databricks with Delta Lake).⁵³ Trino provides full DML (INSERT, UPDATE, DELETE, MERGE) and DDL (CREATE TABLE, ALTER TABLE) support for these formats, enabling true data warehouse-style operations directly on the data lake.^56,⁵⁵}

4.2 Data Mesh

Trino is often cited as the “engine of choice” for implementing a Data Mesh.⁶⁶ A Data Mesh is a decentralized, socio-technical architecture that treats “data as a product.”⁶⁴ In this model, individual business domains (e.g., Marketing, Sales, Finance) are responsible for owning and serving their data products.

Trino’s federation capabilities are a perfect fit for this architecture. Instead of a central ETL team moving all data into one monolithic warehouse, Trino allows a “data consumer” to query data products directly from the domains where they live.⁶³ The Finance team can run a query that federates data from the Sales domain’s lakehouse and the Marketing domain’s CRM database, all without moving the data. This makes Trino a key enabler of the “self-serve data platform” and “federated computational governance” principles of the data mesh.^65,⁶²

4.3 Data Federation vs. Data Virtualization

Trino’s capabilities are often compared to Data Virtualization. While the terms are similar, there is a key distinction:

Data Federation (Trino’s approach): Focuses on providing a unified query layer, but is built as a high-performance MPP engine that assumes the underlying data is “big.” It pushes as much processing as possible (e.g., predicates, aggregations) down to the source systems.^{14,¹⁵,¹⁶}
Data Virtualization (Traditional): This is a broader term for creating a semantic abstraction layer over data.²⁵ While it also federates queries, traditional DV tools often included more robust data management, governance, and caching features, and were not always designed with the same MPP, high-throughput architecture as Trino.^26,²⁷

In practice, Trino is a modern, open-source, high-performance implementation of data federation designed for the data lake and cloud-native era.

5.0 Ecosystem and Competitive Landscape

Trino competes and coexists with a wide array of tools in the data landscape.

5.1 Starburst (Commercial Trino)

Starburst is the commercial enterprise company founded by the creators of Trino.³³

Starburst Enterprise: A supported, enterprise-grade distribution of Trino. It adds features crucial for large organizations, such as enhanced security (global access control, data masking), enterprise-grade connectors (e.g., Teradata, Snowflake), a built-in data catalog, and 24/7 support.³³
Starburst Galaxy: A fully-managed, cloud-native (SaaS) version of Starburst Enterprise, which eliminates the need for organizations to manage their own Trino clusters.³⁴

Starburst’s relationship to Trino is analogous to Databricks’ relationship to Apache Spark or Confluent’s to Apache Kafka. They are the primary commercial backers and contributors to the open-source project.³¹

5.2 Trino vs. Amazon Athena

Amazon Athena is a serverless, interactive query service from AWS.³⁵ It is powered by Trino (and PrestoDB, with AWS migrating to Trino).^40,³⁷

Athena: Easy to use, serverless (no infrastructure to manage), and pay-per-query.³⁶ However, it is a “black box” managed service, can lag behind the latest Trino releases, and has limited federation capabilities outside the AWS ecosystem.
Trino (self-hosted): Offers superior performance (due to fine-tuning), access to the latest features, and true data federation to any source (on-prem, multi-cloud). The trade-off is the operational overhead of managing the cluster.^38,³⁹

5.3 Trino vs. Databricks (Lakehouse Federation)

Databricks, built on Apache Spark, is a primary competitor. In 2023, Databricks announced Lakehouse Federation, a new feature in its Unity Catalog that mimics Trino’s core federation capability.¹⁹

Trino: An open, vendor-neutral query engine designed from the ground up for federation. It is not tied to a specific table format or platform.²²
Lakehouse Federation: A feature of the proprietary Databricks platform. It is designed to pull external data into the Databricks/Spark ecosystem for querying.^{20,²¹ The federation capabilities are newer and less mature than Trino’s extensive connector library.¹⁸}

Many organizations use both: Trino for its broad federation and interactive query strengths, and Databricks/Spark for heavy, large-scale batch ETL and ML workloads.^{47,^24,⁴⁶}

5.4 Other Notable Comparisons

Competitor	Key Differentiator
Google BigQuery	A serverless, proprietary data warehouse.⁴¹ Trino can federate queries to BigQuery, treating it as just another data source, and can be a more cost-effective or performant alternative.^42,⁴³
ClickHouse	An open-source columnar OLAP database.⁴⁴ ClickHouse is a storage system optimized for speed. Trino is a query engine that can query ClickHouse, but its primary strength is federation, not being a standalone database.⁴⁵
Dremio	A direct competitor to Trino/Starburst, also built for lakehouse querying. Dremio heavily emphasizes its “semantic layer” and query acceleration features (e.g., Data Reflections).^48,⁴⁹

6.0 Key Features and Capabilities

Cost-Based Optimizer (CBO): Trino’s CBO analyzes table statistics (row count, nulls, distinct values) from connectors to create more efficient query plans, such as optimizing join order and join distribution types.^51,⁵²
dbt (Data Build Tool) Integration: Trino has a dedicated `dbt-trino` adapter, allowing data teams to use the popular dbt framework to build, test, and deploy data transformation models directly on Trino, leveraging its federation and lakehouse capabilities.⁵⁰
Fault-Tolerant Execution: For long-running, complex ETL queries, Trino can be configured for fault-tolerant execution, which allows it to retry queries or stages upon failure by spilling intermediate data to disk, trading some speed for reliability.⁶¹

7.0 Adoption and Conclusion

Trino is a mature, battle-tested query engine used in production at a massive scale by some of the world’s largest technology companies. The official Trino website lists case studies from companies including Comcast,⁶⁹ Goldman Sachs,⁷⁰ LinkedIn,⁷¹ Lyft,⁷² Netflix,⁷³ Pinterest,⁷⁴ and Salesforce.⁷⁵ Starburst lists further customers including Aljazeera,⁷⁷ Scotiabank, and Slalom.^76,⁷⁸

In conclusion, Trino has established itself as the premier open-source federated query engine for the modern data stack. Its ability to decouple compute from storage and query data in-place makes it a strategic asset for any organization building an open, flexible, and scalable data architecture. By providing a “single pane of glass” for SQL analytics across the entire enterprise—from the data lake to operational databases—Trino effectively breaks down data silos and accelerates time-to-insight. Its strong community, rapid development,^{67,³² and robust backing from Starburst position it for continued growth as the “SQL engine for everything.”²⁸}

Works cited

Apache Trino: Introduction to the Powerful Query Engine – Nitor Infotech, accessed November 7, 2025, https://www.nitorinfotech.com/blog/apache-trino-introduction-to-the-powerful-query-engine/
Trino | Distributed SQL query engine for big data, accessed November 7, 2025, https://trino.io/
History — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/overview/history.html
Announcing Trino | Trino, accessed November 7, 2025, https://trino.io/blog/2020/12/27/announcing-trino.html
Trino: The Origin Story of the New Brand for PrestoSQL – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-the-origin-story-of-the-new-brand-for-prestosql/
What is Trino? The SQL query engine formerly known as PrestoSQL – InfoWorld, accessed November 7, 2025, https://www.infoworld.com/article/3602237/what-is-trino-the-sql-query-engine-formerly-known-as-prestosql.html
Our story | Trino, accessed November 7, 2025, https://trino.io/our-story/
Trino vs. Presto: What’s the Difference? – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/starburst-trino-presto-whats-the-difference/
What is Presto? The SQL engine for big data – InfoWorld, accessed November 7, 2025, https://www.infoworld.com/article/3588923/what-is-presto-the-sql-engine-for-big-data.html
Architecture — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/overview/architecture.html
Trino Architecture – Trino: The Definitive Guide [Book] – O’Reilly, accessed November 7, 2025, https://www.oreilly.com/library/view/trino-the-definitive/9781098103308/ch04.html
Trino Architecture Deep-Dive (Part 1) | by knowledge-center … – Medium, accessed November 7, 2025, https://medium.com/@knowledge.center.b.tech/trino-architecture-deep-dive-part-1-a67b1b01c3d1
Trino concepts — Starburst Enterprise, accessed November 7, 2025, https://docs.starburst.io/latest/overview/concepts/architecture.html
What is Data Federation? | Databricks, accessed November 7, 2025, https://www.databricks.com/discover/data-federation
What is Data Federation? – AWS, accessed November 7, 2025, https://aws.amazon.com/what-is/data-federation/
What is federated data query – Red Hat, accessed November 7, 2025, https://access.redhat.com/documentation/en-us/red_hat_data_services/1/html/get_started_with_querying_across_data_sources/federated-data-query-what-is
Connectors — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/connector.html
Trino vs. Databricks Lakehouse Federation – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-vs-databricks-lakehouse-federation-architecture-and-use-cases/
Introducing Lakehouse Federation in Databricks – Databricks Blog, accessed November 7, 2025, https://www.databricks.com/blog/2023/05/17/introducing-lakehouse-federation-databricks.html
Lakehouse Federation | Databricks, accessed November 7, 2025, https://www.databricks.com/product/lakehouse-federation
Query federation | Databricks Documentation, accessed November 7, 2025, https://docs.databricks.com/en/query-federation/index.html
The Future of the Lakehouse is Federation – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/the-future-of-the-lakehouse-is-federation/
Data Lakehouse vs. Data Federation: Trino vs. Dremio vs. Databricks, accessed November 7, 2025, https://medium.com/@agarti.work/data-lakehouse-vs-data-federation-trino-vs-dremio-vs-databricks-476c1110e58d
Introducing the Trino plugin for Databricks | Trino, accessed November 7, 2025, https://trino.io/blog/2022/05/05/introducing-trino-plugin-for-databricks.html
What is Data Virtualization? – TechTarget, accessed November 7, 2025, https://www.techtarget.com/searchdatamanagement/definition/data-virtualization
Data Virtualization vs. Data Federation: Is There a Difference? – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/data-virtualization-vs-data-federation-is-there-a-difference/
Data Federation vs. Data Virtualization: Key Differences – GSPANN, accessed November 7, 2025, https://www.gspann.com/resources/blogs/data-federation-vs-data-virtualization-key-differences
Use cases | Trino, accessed November 7, 2025, https://trino.io/use-cases/
Trino vs. Presto in 2024: A Tale of Two Open Source Projects – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-v-presto-in-2024-a-tale-of-two-open-source-projects/
What is Trino? The SQL query engine formerly known as PrestoSQL – InfoWorld, accessed November 7, 2025, https://www.infoworld.com/article/3602237/what-is-trino-the-sql-query-engine-formerly-known-as-prestosql.html
Trino: The Definitive Guide (O’Reilly Book) – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-the-definitive-guide-oreilly-book/
Trino Summit 2024 recap | Trino, accessed November 7, 2025, https://trino.io/blog/2024/02/02/trino-summit-2024-recap
Starburst | The Data Lakehouse Analytics Platform, accessed November 7, 2025, https://www.starburst.io/
Starburst Galaxy | Fully-Managed Data Lakehouse Platform, accessed November 7, 2025, https://www.starburst.io/platform/starburst-galaxy/
Amazon Athena | Serverless Interactive Query Service – AWS, accessed November 7, 2025, https://aws.amazon.com/athena/
Amazon Athena Pricing – AWS, accessed November 7, 2025, https://aws.amazon.com/athena/pricing/
Amazon Athena FAQs – AWS, accessed November 7, 2025, https://aws.amazon.com/athena/faqs/
Trino vs. Athena: Which is Better and Why? – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-vs-athena-which-is-better-and-why/
Trino vs. Athena: A Practitioner’s Guide – Invisory, accessed November 7, 2025, https://www.invisory.io/blog/trino-vs-athena
Trino vs. Athena | Trino, accessed November 7, 2025, https://trino.io/blog/2023/10/26/trino-vs-athena
BigQuery: Cloud Data Warehouse – Google Cloud, accessed November 7, 2025, https://cloud.google.com/bigquery
Trino vs. BigQuery: 7 Key Differences – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-vs-bigquery/
BigQuery connector — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/connector/bigquery.html
ClickHouse | Fast Open-Source OLAP DBMS, accessed November 7, 2025, https://clickhouse.com/
Trino vs. ClickHouse | Trino, accessed November 7, 2025, https://trino.io/blog/2023/01/19/trino-vs-clickhouse
Trino vs. Databricks SQL: 5 Key Differences – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-vs-databricks/
Trino and Databricks: Better Together – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-and-databricks-better-together/
Trino vs. Dremio: 5 Key Differences – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-vs-dremio-5-key-differences/
Dremio vs. Trino | G2, accessed November 7, 2025, https://www.g2.com/compare/dremio-vs-trino
Trino and dbt | Trino, accessed November 7, 2025, https://trino.io/blog/2024/05/23/trino-and-dbt
Cost-based optimizer — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/optimizer/cost-based-optimizer.html
Cost-Based Optimizer | Trino, accessed November 7, 2025, https://trino.io/blog/2017/05/15/cost-based-optimizer.html
Apache Iceberg in Trino | Trino, accessed November 7, 2025, https://trino.io/blog/2023/10/18/apache-iceberg-in-trino
Iceberg connector — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/connector/iceberg.html
Run Trino queries on Apache Iceberg tables in AWS Lake Formation – AWS, accessed November 7, 2025, https://aws.amazon.com/blogs/big-data/run-trino-queries-on-apache-iceberg-tables-in-aws-lake-formation/
Trino & Iceberg: Powering the Open Lakehouse – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/trino-iceberg-powering-the-open-lakehouse/
What is a Data Lakehouse? – Starburst, accessed November 7, 2025, https://www.starburst.io/learn/data-fundamentals/what-is-a-data-lakehouse/
Iceberg, Delta Lake, Hudi: Lakehouse table formats and Trino | Trino, accessed November 7, 2025, https://trino.io/blog/2023/07/26/iceberg-delta-lake-hudi-lakehouse-trino
Delta Lake connector — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/connector/delta-lake.html
Hudi connector — Trino 450 Documentation, accessed November 7, 2025, https://trino.io/docs/current/connector/hudi.html
Trino | Episode 53 – Fault-tolerant execution in Trino, accessed November 7, 2025, https://trino.io/episodes/53
Trino | Episode 31 – Data mesh with Starburst, accessed November 7, 2025, https://trino.io/episodes/31
Data Mesh: The What, The Why And The How – Forbes, accessed November 7, 2025, https://www.forbes.com/sites/forbestechcouncil/2022/08/23/data-mesh-the-what-the-why-and-the-how/
How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh, accessed November 7, 2025, https://martinfowler.com/articles/data-mesh-principles.html
What is Data Mesh? – Starburst, accessed November 7, 2025, https://www.starburst.io/learn/data-mesh/what-is-data-mesh/
How Starburst and Trino Power a Data Mesh – Starburst, accessed November 7, 2025, https://www.starburst.io/blog/how-starburst-and-trino-power-a-data-mesh/
The state of Trino | Trino, accessed November 7, 2025, https://trino.io/blog/2022/11/17/state-of-trino
Case studies | Trino, accessed November 7, 2025, https://trino.io/case-studies/
Comcast case study | Trino, accessed November 7, 2025, https://trino.io/case-studies/comcast.html
Goldman Sachs case study | Trino, accessed November 7, 2025, https://trino.io/case-studies/goldman-sachs.html
LinkedIn case study | Trino, accessed November 7, 2025, https://trino.io/case-studies/linkedin.html
Lyft case study | Trino, accessed November 7, 2025, https://trino.io/case-studies/lyft.html
Netflix case study | Trino, accessed November 7, 2025, https://trino.io/case-studies/netflix.html
Pinterest case study | Trino, accessed November 7, 2025, https://trino.io/case-studies/pinterest.html
Salesforce case study | Trino, accessed November 7, 2025, https://trino.io/case-studies/salesforce.html
Our Customers – Starburst, accessed November 7, 2025, https://www.starburst.io/customers/
Al Jazeera | Breaking News, World News and Video from … – Al Jazeera, accessed November 7, 2025, https://www.aljazeera.com/
Slalom | Global consulting firm focused on strategy, technology …, accessed November 7, 2025, https://www.slalom.com/

A Strategic Report on Presto, Trino, and Starburst