Presented by:

Ning Yu

Beijing Siwei Zongheng Data Technology Co., Ltd.

Ning Yu is a database systems expert specializing in database performance optimization, storage engine design, and vectorized execution engines. He is a core contributor to the Greenplum kernel development team and is ranked among the top five global contributors to the Greenplum project. In addition, Yu is an experienced Linux kernel expert. His work spans operating system internals and high-performance database execution architectures, bridging low-level system design with modern data infrastructure innovation.

Yandong Yao

Beijing Siwei Zongheng Data Technology Co., Ltd.

Founder & CEO, Beijing Siwei Zongheng Data Technology Co., Ltd. Former General Manager, Greenplum Beijing R&D Center Yan-Dong Yao is a pioneer in open-source data technologies and ecosystem development. He founded the Greenplum China Open Source Community while serving on the standing committee of the PostgreSQL Chinese Community. He is also a co-founder of the 1024 Digital Foundation. Since Greenplum’s open-source release, Yan-Dong has led efforts to grow its ecosystem in China, establishing a significant user base across major cloud providers including Alibaba Cloud, Tencent Cloud, and Baidu Cloud. He holds multiple domestic and international patents in cloud computing and big data, and authored Greenplum: From Big Data Strategy to Implementation. Before his work with Greenplum, he worked in systems and storage at Sun Microsystems and Symantec. Yan-Dong graduated with honors from the Institute of Software, Chinese Academy of Sciences.

No video of the event yet, sorry!

PostgreSQL's performance in OLAP and HTAP scenarios can be significantly enhanced through a vectorized executor, an optimized execution model tailored for analytical workloads. By leveraging typed operations, batched processing, and SIMD instructions, vectorized execution minimizes overheads associated with traditional row-by-row processing, such as repeated function calls and data extraction from tuples. This talk explores how PostgreSQL’s existing extensibility mechanisms—including slots, custom scans, planner hooks, and table AMs—enable the development of a pluggable vectorized executor without modifying core kernel code. Key design considerations include: 1. Columnar Data Support : Utilizing columnar storage to reduce I/O and improve cache locality, with benefits like pre-sorting, compression, and predicate pushdown. 2. Batched & Vectorized Operations : Processing data in batches to leverage CPU cache efficiency and SIMD instructions, reducing per-tuple overhead. 3. Seamless Integration : Ensuring coexistence with PostgreSQL’s traditional executor via custom scan nodes and planner hooks, enabling dynamic plan selection based on workload type. Performance benchmarks will demonstrate speedups over the standard executor, with perf profiling highlighting optimizations like reduced function call overhead and improved instruction pipelining. Future directions include enhancing memory management and parallel execution to further boost OLAP performance while preserving PostgreSQL’s OLTP capabilities.

Date:
2026 April 21 11:00 PDT
Duration:
20 min
Room:
San Pedro (Level C)
Conference:
Postgres Conference: 2026
Language:
Track:
Dev
Difficulty:
Medium