blog/Tail Latency & System Behavior

Topic · 6 posts

Tail Latency & System Behavior

Runnable Java experiments on the failure patterns that show up under real load: tail latency, queueing, hedged requests, coordinated omission, backpressure, and SLO engineering. Deterministic outputs, checked-in CSVs, reproducible on any machine.

posts6

first post2026-05-31

latest2026-06-28

total read~46 min

All Project Loom Structured Concurrency

6 posts

newest first

012026-06-28
SLO Engineering & Designing Predictable Services Under Load
A deterministic Java SLO simulation against a 99%-under-200ms objective. A timeout caps p99 at 200ms yet scores the same 57.36% SLI as doing nothing, because a fast failure is still a failure; a bulkhead holds 99.60% with 60% of the error budget left. Burn rate fires from a single window.
7 min
022026-06-28
Backpressure Design Patterns
A deterministic Java comparison of five admission strategies under identical 2x overload. Reject-fast policies (token bucket, rate limiter) hold p99 at 10ms; a bounded queue accepts fewer requests at a 500ms p99, because under sustained overload a buffer is the worst of both.
7 min
032026-06-21
The Coordinated Omission Problem
A deterministic Java measurement harness where a closed-loop benchmark reports a whole-run p99 of 10ms for a service that froze for 500ms, while open-loop scheduling and HdrHistogram's coordinated-omission correction both report 460ms from the same samples.
6 min
042026-06-14
Hedged Requests & Speculative Execution
A deterministic Java hedging simulation where a p95 threshold cuts p99 from 200ms to 43ms for 3.7% extra load, while a p99 threshold improves nothing and burns the most wasted work. Hedging is a latency technique you buy with capacity.
7 min
052026-06-07
Queueing Theory for Engineers
A deterministic Java queueing simulation where p99 stays at 10ms through rho=1.00, then jumps to 109ms at rho=1.05 and 605ms at rho=1.30 while service time stays fixed.
10 min
062026-05-31
Why Average Latency Lies
A deterministic Java simulation where baseline p99=34ms becomes fan-out p99=597ms without changing the downstream latency sampler. Average latency is structurally incapable of catching that tail pain.
9 min