kPacketInterval * i with a size_t i gives the deadline an unsigned
duration; once the producer falls behind, libc++'s sleep_until turns the
negative remaining time into a near-infinite sleep and the benchmark
hangs. Multiply by a signed index.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A hidden Catch2 case ([.benchmark], run with ./tests "[benchmark]") that
pushes 1000 synthetic H.264-like packets (150 KB keyframe every 50, 15 KB
deltas) through a StreamSocket at 250 packets a second with 0, 1 and 8
consumers, and prints per scenario: SendMedia wall time on the producer
thread (p50/p99/max), CPU per packet for the producer, the listener
thread and an average consumer, and how many packets each consumer
received. Producer CPU is sampled around each SendMedia with the clock
read overhead calibrated out; the listener figure is the process CPU
left after subtracting the producer thread and the consumers.
Run on a 4-core VM after the unlocked drain change: about 1.3 us per
SendMedia with no consumer, 16 us with one, 18 to 24 us with eight,
and every consumer receives every packet.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
