Course logistics

  • Homework 2 released, due Sunday September 22 — three multiple-choice questions, submitted in Gradescope (0:27).
  • Project 1 (buffer pool manager) is live: write-up on the website, code and Piazza announcement posted; recitation Wednesday 6 p.m. (0:27).
  • Advice: read the Project 1 write-up beforehand so recitation questions are about approach, not "what's a buffer manager" (1:14).
  • The buffer pool manager lecture lands Monday (1:14); industry-partner visit day is Monday/Tuesday with research talks and info sessions, including Confluent (1:46).

Where this lecture sits

  • This is the last lecture on the storage layer — the bottom layer of the DBMS (0:27).
  • Prior lectures covered write-optimized approaches (2:25):
    • Tuple-oriented storage with slotted pages.
    • Log-structured storage — append changes; writes get faster, reads get slower.
    • Index-organized storage — a B+Tree whose leaf nodes hold the data itself.
  • Those approaches serve write-heavy workloads (inserts/updates/deletes/upserts) (3:07).
  • Today's target is the opposite: read-only analytical queries where SELECT speed matters, not insert speed (3:34).

Agenda

  1. Categorize database workloads and fix the vocabulary for the rest of the semester (4:05).
  2. Storage models — how tuples are physically represented (4:20).
  3. Compression, which the column-store layout unlocks (4:30).
  4. Guest talk from StarTree, an analytical system built on the columnar approach (4:30).