Following Google or Don't Follow the Followers, Follow the Leader
summary
Argues that real decisions come from scale not buzzwords, showing why Google’s stack matters: Megastore → Dremel → Spanner → F1 delivers global ACID transactions, protobuf columns and ~200 ms user roundtrips. Contrasts approaches with concrete tradeoffs: sharding, memcached, ETL, OLAP and Hadoop’s batch limits; cites eBay’s 40 petabytes log store and Microsoft’s median analytic size of 14 GB to show when relational MPP, NoSQL or Hadoop is actually the right tool.
AI-written summary · about
chapters (6)
- 0:00Start
- 8:46More information = need for new metadata techniques: content tagging, author catalogs
- 10:19Paper Tech v2.1: Increased storage density, smaller form factor, durability, high res RGB graphics
- 19:56Information management through human history always follows the same pattern
- 40:34Relational: a good conceptual model, but a prematurely standardized implementation
- 43:02Pricing and performance reality: Hadoop is a storage play, not a database of analytics play
related talks
30mArchitecture & Teams at Less-than-Google Scale
53mWhat I Learned About Going Fast at eBay & Google
41mThe monolith must die! - How to convince managers
52mThe Future of Software Engineering
18mMoving Fast at Scale
11mStaying Relevant in an Ever-changing Tech Landscape
12mDigital & Data/AI Strategy and Transformation
11mScaling systems, scaling influence: Lessons from distributed architecture