Above the Clouds: Investigating Systems from 32,000 ft.
summary
Three concrete case studies: load-testing with the Erlang tool Bench, a quick Bash hammer to spawn many Erlang VM workers, and a discrete-event simulator in Ruby called dsq. A seven-node react cluster runs locally, gets pounded and one node is killed to observe failover; metrics and ubi‑graph visuals show spikes in disk I/O and throughput drops. The surprising result: request timing variability, not too few Mongrels, makes Postgres choke, and boosting Postgres CPU/RAM fixes spikes more cheaply than adding app workers.
AI-written summary · about
chapters (10)
- 0:00Start
- 0:06Above the Clouds
- 1:31The Opportunity The Problem
- 7:29DSQ Darn Simple Queuing
- 10:37Simulating Systems is Hard
- 10:53Simulating Distributed Systems is Distributedly Hard
- 13:46Benchmark
- 16:47Monitoring
- 26:18Cluster Trouble
- 28:38Reminer: The Point
