Arrow Flight SQL Server

Performance Benchmark Summary and Sizing Recommendations

This paper presents benchmarking results and sizing guidance for the Teradata Fabric Arrow connector, which implements Apache Arrow Flight SQL, a modern protocol for high-speed data transfer between SQL databases and clients. Apache Arrow provides an in-memory columnar format, extended by Arrow Flight to enable parallel data streaming over gRPC, and further extended by Arrow Flight SQL to apply this capability to SQL database interactions. This stands in contrast to the long-dominant JDBC protocol, which is limited by its single-threaded design for data reads.

The core focus of the benchmarks is comparing data read (download) performance between traditional JDBC and the Teradata Arrow Flight SQL Server, testing both single-stream and multi-stream Arrow Flight SQL configurations. Because the Teradata Arrow Flight SQL Server can deliver data in parallel columnar gRPC streams, reads are dramatically faster than what JDBC can achieve. The paper also covers upload performance comparisons between JDBC and Arrow Flight SQL's DoPut capability.