Introducing Prestissimo, Presto's Next-Gen Query Execution Engine
summary
Prestissimo is a drop-in C++ worker for Presto that replaces the Java worker and runs plans over the Velox vectorized runtime. Benchmarks on Parquet data in S3 (TPCH, warmed runs) show up to 4.3x throughput, ~2.6x faster elapsed time and ~3x lower CPU, enabling roughly 1/3 the prior CPU for the same workload. Velox provides SIMD vector processing, adaptive hash strategies, a spiller and memory arbitration, and an async read/cache layer that together remove Java GC cliffs and improve operational predictability.
AI-written summary · about
chapters (9)
- 0:00Prestissimo and Presto intro
- 5:21Presto architecture overview
- 9:54Why Prestissimo C++
- 13:49Prestissimo architecture benefits
- 16:59Velox library role
- 23:43Velox vectorized execution
- 26:53Velox reusable library
- 31:04Benchmarking performance
- 33:11Current status future
related talks
22mIntroduction to Presto: The SQL Engine for Data Platform Teams
22mPresto for Your Fast, Open SQL Lakehouse
49mPresto in the Cloud
50mPresto: SQL-on-Anything using Kubernetes
17mComposable Data Management Systems
34mBuilding Presto SQL experience on Flink
28mOptimizing Latency-Sensitive Queries for Presto at Facebook
48mContinuous Optimization for distributed BigData analysis