Simulant in Anger
summary
Simulation testing with Simulant models multi-actor mobile-wallet flows, samples many user sequences, and replays logged actions for retroactive validations. Building a smoke sim takes about a week, a full multi-user sim about a month; fixes after simming take one to two weeks. Simulation uncovers that the system is I/O bound (auto-scaling triggered on CPU fails), explains a 75% error rate, and after replacing auth calls with JWTs throughput jumps from 30→170 tpm while mean latency halves and errors drop to zero.
AI-written summary · about
chapters (22)
- 0:00Simulant in Anger An Experience Report
- 2:33Landing on Mars
- 3:08Big Launch 100s Stores 1000s of Tellers
- 3:22Big Launch No Second Chances
- 5:22Simulating Reality
- 6:21Enter Simulation Testing
- 8:37Enter Simulant
- 13:19Step One Goals
- 13:41Incrementally Improve
- 15:32And what did we find...
- 18:40JWTS Encrypted Payloads As a Token
- 19:43Mean Response Time 2x Improvement
- 20:46Overall 4.5x Throughput 30 TPM Constant 170 TPM Proportional
- 21:38Eroded False Confidences
- 22:05Gained True Confidence
- 22:22A Tool For Change
- 22:55Fantastic for Qualities
- 23:20Is Simulation Testing for you?
- 24:10High Risk necessitates High Confidence
- 24:27Reputation Money Lives...
- 24:49Moderate Risk necessitates High Confidence
- 25:11Unsession!
related talks
42mSimulation Testing
40mTesting Distributed Systems w/ Deterministic Simulation
25mBeware the Bot Armies! Protecting RTC User Experience Using Simulations
10mInnovate Live Streaming with a Client Simulator
01h 52mIOSim (Advanced Track)
30mBreak the API Test Data Bottleneck: Real Scenarios on Real Systems
08mVariations During Testing
23mSim-ln