Scalability Is Quantifiable: The Universal Scalability Law
summary
Claim: scalability is a simple quantitative function. The Universal Scalability Law models throughput vs concurrency with two measurable parameters: sigma (serialization) and kappa (crosstalk). Fit the USL to ordinary monitoring data (concurrency and throughput) with regression, forecast beyond observed load, and identify whether limits come from serialization or quadratic coordination costs that cause retrograde scalability and thrashing. Examples shown include mutex wake‑up storms in MySQL, benchmark comparisons (Java vs Node) and team‑size overheads (Brooks / two‑pizza).
AI-written summary · about
chapters (22)
- 0:00Intro
- 0:37How Systems Fail Under Load
- 1:24Failure Boundaries
- 2:42Workload Failure Isn't Crisp
- 4:12The Failure Boundary Is Nonlinear
- 5:05Capacity
- 5:56Queueing Theory
- 8:21Scaling A System: Ideal
- 10:24Equation
- 12:13Amdahl's Law
- 13:04But What If Workers Coordinate?
- 14:17N Workers = N(N-1) Pairs
- 14:44Universal Scalability Law
- 17:06Crosstalk Penalty Grows Fast
- 18:25More About Crosstalk
- 19:47How Do You Measure Parameters?
- 22:36What is Scalability?
- 23:22But What Is Load?
- 23:41Forecast Workload Failure Boundary
- 25:36Characterize Non-Scalability
- 26:26How Scalable SHOULD It Be?
- 28:13See Your Teams As Systems
related talks
43mThe Math of Scalability
45mThe Universal Laws of Performance
46mPerfect Scalability
38mThe SRE’s Crystal Ball: Predicting System Performance with...
35mSystems Scale Summer Q&A Week 1
29mScaleCheck: A Single-Machine Approach for Discovering Scalability Bugs...
38mHow Scale Makes Distributed Systems Slower
32mAn Uncoordinated Approach To Scaling (Almost) Infinitely