What a DBMS gives you
- Software that lets applications store, analyze, and query data in a database (
16:00). - A general-purpose DBMS supports arbitrary schemas and arbitrary queries — not hardcoded to one lookup (
16:00). - Database systems are among the most widely deployed, tested, and vetted software in the world — rivaled only by operating systems, with multi-billion-dollar companies riding on their correctness (
16:30). - Advice: never build your own DBMS inside your application. Use an existing one (
17:09).
Data model vs. schema
- Data model — a high-level abstraction for defining collections of data and the relationships between them (
17:31,19:20). - Schema — within a chosen data model, the specific tables/collections, their attributes, and their types (
18:12). - Without a schema the data is "a bunch of random bits" and is meaningless — the schema is what makes queries possible (
18:12).- Log-file example: raw text lines are unqueryable; declaring
timestamp → event code → messageturns them into something you can query (18:41).
- Log-file example: raw text lines are unqueryable; declaring
- Most DBMSs support exactly one data model, but the boundary is now blurry: relational databases store JSON, and document databases have added SQL (
17:47).
Catalogue of data models
| Model | Examples / notes |
|---|---|
| Relational | The course's focus; most real systems (20:10) |
| Key/value | RocksDB, LevelDB, Redis — value has no meaning beyond what the app imposes; used for simple apps and caching (20:53) |
| Document / JSON | MongoDB (21:31) |
| Wide column / column family | A subset of document DBs — Google BigTable, Cassandra (21:31) |
| Array / matrix / vector / tensor | Tensors are the one thing awkward in the relational model; the SQL standard only added multidimensional arrays recently (21:31) |
| Hierarchical, network, semantic, entity-relationship | 1970s–80s experiments — "they were doing cocaine back then" (22:13) |
The "NoSQL" arc
- NoSQL originally meant "no SQL"; it now means "not only SQL" — the camp came back around to relational ideas (
20:53). - The claimed 10-year cycle: someone declares the relational model and SQL obsolete, Hacker News gets excited, and eventually everyone concedes the relational model was right (
22:13). - Prediction stated in lecture: vector databases are the current iteration and will also come around to SQL (
22:38).