juhapekkaeloranta / data-access-patterns

Ideas/notes for a masters thesis

data-access-patterns

• How many values accessed?

  • Single-value (result single row)
    • 0-1
  • Multiple-values
    • 0-n

• Crud-type?

  • read-only SELECT ...
  • append-only INSERT INTO ...
  • edit UPDATE VALUES ...
  • delete DELETE FROM ...

• CRUD-types -> frequency of each type

  • see https://gpdb.docs.pivotal.io/43180/admin_guide/ddl/ddl-storage.html

• Analysis queries

  • Redis vs materialized view vs columnar vs hadoop/etc
  • Predictability (cubes, rollups, how detailed?)

• OLTP / OLAP

• Read-order/area

  • sequential / random
  • sparse / tight
  • rows / disk-block ratio?
  • predictability? (think document vs normalized)
  • swapping rate

• Query size

  • Query span fits in memory?
  • Query result fits in memory?

• Isolation-requirement

  • isolation levels
  • optimistic / pessimistic locking!
  • event time? start / end of transaction

• Durability requirements?

• Recency level?

  • eventual consistency ok?

• Main-memory heavy operations?

  • sort
  • aggregates (?)
  • joins

• How much data?

• How many accessors? (appliocation replicas)

  • hotspots? vs easy sharding?

• CPU usage?  - where condition evaluation cost?

• Latency requirements

• "Programming with read-only db"

• Immutability

  • temporal data

• Event-driven database

  • see http://dchambers.github.io/articles/databases-are-dead/

• Time-series data

• Insert data clustering

  • Think citybikes