Breaking the Memory Wall in MonetDB

Boncz, Kersten, Manegold

rdbms database architecture column store cache tuning algorithms memory

@article{boncz:cacm-2008,
  title={Breaking the Memory Wall in {MonetDB}},
  author={Peter A. Boncz and Martin L. Kersten and Stefan Manegold},
  journal={Communications of the {ACM}},
  volume={51},
  number={12},
  month={December},
  year={2008}
}

Database technology is increasingly outdated

• Originally designed for architectures in which disk I/O was the dominating performance factor
• Originally designed for transaction processing

Computer architecture has changed dramatically though

• Multiple levels of cache must be effectively utilized, for both data and instructions
• Processor pipelining requires independent, predictable code

Database usage has also evolved from transaction-oriented use to include online analysis of large data sets

MonetDB covers several areas to address these:

• Storage based on columns, which better utilizing caches, particularly for analysis tasks
• Simplified, highly optimized, and cache conscious query algebra and operator algorithms, implementations
• Self-calibrating, optimizing query path planning based on sophisticated memory access cost models