EDB, a prominent player in the PostgreSQL ecosystem, has unveiled enhancements to its data platform, EDB Postgres AI, designed to integrate transactional, analytical, and AI workloads within a unified environment. This innovative release introduces a specialized engine for PostgreSQL that boasts independent scalability from cloud storage and is finely tuned for columnar open table formats such as Iceberg and Delta Lake.
Performance Enhancements
According to EDB, this new engine leverages open source Apache DataFusion to execute queries with remarkable efficiency, achieving speeds up to 30 times faster than traditional PostgreSQL. Additionally, the tiered storage solution is reported to offload cold transactional data to storage at an impressive 18 times greater cost efficiency.
DataFusion, an extensible query engine crafted in Rust, utilizes Apache Arrow as its in-memory format, enhancing the overall performance of data operations.
Tom Kincaid, EDB’s senior vice president for database server and tools, elaborated on the capabilities of the analytics accelerator: “Our analytics accelerator is a tiered table, where particular database analytical queries can be pushed to another open source engine, Apache DataFusion. We’ve done the glue and the hook-up between that and our Postgres distributed product to make an analytics accelerator that you can use right from within PostgreSQL.”
As part of the EDB Postgres AI platform, this tiered storage and query engine promises a total cost of ownership (TCO) that is six times more favorable and transactional performance that is 30 percent faster than SQL Server, according to the vendor’s claims.
Shifting Paradigms in PostgreSQL
Traditionally recognized as a transactional database, PostgreSQL is now evolving to accommodate a growing demand for analytics within the same environment, particularly to support AI applications. This shift follows EDB’s strategic decision to fork Greenplum, an OLAP system built on PostgreSQL, to create WarehousePG.
“We’ve also started the open source data warehouse PostgreSQL project, which effectively uses Greenplum-related workloads doing traditional analytical queries,” Kincaid noted, drawing on his previous experience as an engineering vice president at Oracle.
However, EDB’s goal is not to redefine PostgreSQL as an analytics database but rather to provide users with the tools necessary to conduct analytics seamlessly within the PostgreSQL framework. Kincaid explained, “The threshold for when you need to go from a traditional PostgreSQL, OLTP database to a specialized analytics variant will get a little higher. Maybe it will be set around the 2TB range now, and maybe it goes up to the 3TB range now.”
While PostgreSQL will continue to serve as an OLTP database for a significant portion of use cases, it is increasingly equipped to handle analytical workloads alongside transactional processing. Kincaid concluded, “I wouldn’t say it will be a full-scale analytics engine in the future.”
