Following Google or Don't Follow the Followers, Follow the Leader
summary
Argues that following Google research beats following vendors or Yahoo and that MapReduce is now a decade‑old core (Hadoop inherits it) and is largely obsolete for many workloads. Names concrete tradeoffs: scale, speed, persistence, context; cites a Microsoft study showing median analytic job input ~14GB; points to Google papers Spanner/F1 (2012–2013) combining ACID, cross‑datacenter sync replication and SQL over distributed storage as the direction to emulate rather than shoehorning applications into ad‑hoc NoSQL stacks.
AI-written summary · about
chapters (51)
- 0:00Intro
- 2:13Early history
- 10:09Scrolls
- 12:06Clay vs Paper
- 12:55Codex
- 13:44Medium encyclopedias
- 14:10Movable type
- 16:02Perfect copies
- 17:31Metadata
- 18:25Sharing knowledge
- 19:23Taxonomy
- 20:00Species are not fixed
- 21:50Victorian Era
- 22:24Industrial Era
- 24:01Topdown Taxonomy
- 25:16Why did this happen
- 25:35Problems of pragmatism
- 27:57Scale changes
- 28:56Top down vs bottom up
- 30:20Scale changes architectures
- 31:19Big data
- 32:05Relational databases
- 32:35Moores law
- 33:17Clientserver
- 33:49Connection pooling
- 34:46Stateless architectures
- 35:11Database scalability
- 36:02Sharding
- 37:15Design for failure
- 39:09You can put this new thing in
- 39:29So what have we done
- 40:04Big mainframes
- 40:15Data warehousing
- 41:47Cubing architectures
- 42:39Database architectures
- 43:12No sequel
- 44:42Query optimization
- 44:58Data modeling
- 47:17Performance curve
- 48:06Key value store
- 48:18Query strings
- 49:23Hadoop
- 49:49Microsoft
- 50:29Model building
- 50:56Unstructured data
- 51:39Scientific computing
- 52:11Query problems
- 52:26Small analytics
- 53:24Hadoop and no sequel
- 54:09Big conclusion
- 54:32F1 evolution
