Parallel Execution — References

Concurrency structure, the latency arithmetic of fan-out, and the tracing needed to see inside it.

Structured concurrency

• Python asyncio task groups — scoped concurrent tasks with cancellation on failure and guaranteed completion before the scope exits.
• Trio — nurseries as an enforced scope for concurrent tasks; the reference implementation of the idea that a function must not outlive its callees.
• Notes on structured concurrency, or: Go statement considered harmful (Nathaniel J. Smith,
  2018. — the argument that unscoped task spawning is the concurrency equivalent of an unrestricted jump. Cited by name.

Latency under fan-out

• The Tail at Scale (Jeffrey Dean and Luiz André Barroso, 2013) — why tail latency dominates any operation that waits on many components, and the hedged-request and tied-request techniques for containing it. Cited by name.
• Amdahl's law — the ceiling on speedup imposed by whatever fraction of the work remains sequential, which sets a hard limit on what fan-out can achieve. Cited by name.

Seeing inside concurrent work

• OpenTelemetry — span relationships and links, which are what make a fan-out readable as a tree rather than an interleaved stream of events.

No verified reference is recorded for concurrency limits specific to model-provider rate limiting; the operative constraints are per-account quotas documented by each provider and deliberately not reproduced here.