The Blaze project aims to provide Spark with a high-performance, low-cost native execution layer.
We seek to solve a series of performance bottlenecks in the current JVM-based Task execution of Spark SQL, such as high fluctuations in performance due to GC, high memory overhead, and inability to accelerate computation directly with SIMD instructions.
This repo is under active development and is not ready for production (or even development) use, but stay tuned for updates!
A work-in-progress to catalog the state of support for TPC-DS.
Please click on the ❌ next to the commit ID to check how far we are 😉.