The following are the learnings from the podcast:

  • Indexing was the realm of DBA experts
  • Machine learning to learn to indexing
  • Simple ML algos to complex algos are being used to learn the index maps
  • Btrees, Hashmaps, Bloomfilters - typical algos to build indices
  • DBA will need to become machine learning experts
  • Many areas such as query optimisation, multivariate indexing learning models, indexing for inserts are all the exciting new development s waiting to happen
  • Data distribution changes- custom indexing atrategies do not work anymore
  • Use GPU to train models
  • Indexes- used for preprocessing
    • Btrees - Range of records
    • Hashmaps- Specific queries
    • Bloom filters- efficient storage
  • Change in complexity class for a specific kind of indexing
  • Log(n) space to constant time
  • Btrees speed up 2x - Indexes significant reduction in space
  • Supervised learning
  • Latency in nanoseconds
  • Redix Sort
  • Training time is dependent on the algo