Practical performance theory

on YouTube ↗06m

" summary

Models treat the web server as an M/G/1-style queue with assumptions: independent random arrivals, FIFO single-threaded processing, and constant service time with no downstream bottlenecks. Utilization follows utilization = arrival rate × service time, queuing delay scales like u/(1-u) so response time hockey-sticks as u approaches 1, and the practical capacity bound hits about ~5500 req/s where the response-time SLA is breached.

AI-written summary · about

" chapters (4)

1. 0:00Start
2. 0:05performance modeling
3. 1:21model the web server as a queueing system.
4. 2:52"What's the maximum throughput of this server?" le given a response time target

" related talks

• 38mA Practical Look at Performance TheoryKavya Joshi · Strange Loop 2018
• 46mQueueing Theory in Practice: Performance Modeling for the Working EngineerEben Freeman · LISA17
• 30mQueueing Theory, In Practice: Performance Modelling in Cloud-Native Territory [I]Eben Freeman · KubeCon + CloudNativeCon North America 2017
• 47mPractical Performance AnalysisSimone Bordet · Voxxed Days Ticino 2024
• 42mStop Rate Limiting! Capacity Management Done RightJon Moore · Strange Loop 2017
• 45mThe Universal Laws of PerformanceRaphaël Luta · Devoxx France 2022
• 45mUnderstanding Microservice PerformanceRob Harrop · Spring I/O 2016
• 30mWhat's the Cost of a Millisecond?Avishai Ish Shalom · SREcon21