Advanced Napkin Math: Estimating System Performance from First Principles
Talkat SREcon19 Europe/Middle East/Africa · on Thursday, 2019-10-03· at 09:00· Track 1: The Liffey B·
summary
Shows how to turn guesses into numbers: memorize base rates (network, SSD, RAM, CPU at $10 per CPU) and combine them to decide if designs work. Measured examples include an 8‑byte Redis GET at about 25 microseconds (people guess ~100µs) and dumping 16 GB across regions taking on the order of 15 minutes—so naive snapshot‑and‑restore failover is infeasible. Uses Fermi decompositions, sequential vs random I/O rates, and inverted‑index sizing to pick practical architectures.
AI-written summary · about
chapters (22)
- 0:00Introduction
- 6:26Cloud bill example
- 7:15Wheres the other 90
- 7:41Base Rates
- 9:29Eratosthenes
- 12:35Chicago
- 16:08Problem
- 18:39Big O notation
- 20:19Failover numbers
- 22:08Mailing list
- 22:35Existing system
- 23:52TCP
- 24:12HTTP
- 25:13Shopify
- 27:41Decision Tree
- 28:04Napkin Math
- 28:52Inverted Index
- 31:49Memorization
- 32:26Space Repetition
- 32:56Periodic Challenges
- 33:22Applying Napkin Math
- 33:57Questions
related talks
42mAdventures in Efficiency and Performance
01h 04mUnlocking Performance Through Reverse Engineering
01h 00m“When a Microsecond Is an Eternity: High Performance Trading Systems in C++”
51mDesigning for Performance
56mWrite Fast Code Like a Native
37mCausal's 1000x Spreadsheet; Performance beyond profiling
01h 50mUnlocking Modern CPU Power - Next-Gen C++ Optimization Techniques
50mEnergy Efficient Design Principles of Software Engineering