Learning series · 4 episodes · Staff / Senior
Where the time goes
Five engineering questions about network latency, load distribution, constrained concurrency, queueing, and retry overload.
Start series →Follow the decisions in order
- Part 1 of 5
Staff
The server finished in 30 ms. Where did the other 370 ms go?Separating end-to-end client latency from instrumented server work, including requests that never produce an application span.
- Part 2 of 5
Staff
We added six servers. Why did none get traffic?The difference between assigning transport connections and routing requests when HTTP/2 channels are long-lived.
- Part 3 of 5
Senior
CPU is at 20 percent. Why are 12,000 requests waiting?Using Little’s Law and pool budgets to explain why low CPU does not mean spare request capacity.
- Part 4 of 5
Staff
Traffic rose 15 percent. Why did latency jump sixfold?Understanding nonlinear queueing delay near saturation without turning a simplified model into a universal rule.
- Part 5 of 5
Staff · AI Engineering reading
How can three retries turn a small model outage into a platform incident?Removing nested retry amplification, bounding attempts and admitted load, and controlling recovery.