Optimizing the computation (cont.)

• Cache results

  • caching (in memory) the results of a group-by from which other group-byes are computed (less I/O)

• Amortize-scan

  • amortizing disk reads by computing as many group-bys together

• Share-sorts (for SORT only)

  • sharing sorting cost across multiple group-bys

• Share-partitions (for HASH only)

  • when the hash table is too large, data is partitioned and aggregation is done for each partition that fits in memory

THEY REPRESENT TRADE-OFFS

Previous slide

Next slide

Back to first slide

View graphic version