bench

Benchmark Results

CZMQ 4.2.1 | ZMQ 4.3.5 | Ruby 4.0.1+YJIT | Linux x86_64

Throughput (push/pull, iterations/s)

Async (fibers)

Size	inproc	ipc	tcp
64B	283.6k	17.3k	11.4k
256B	292.0k	17.1k	18.6k
1024B	251.7k	16.7k	13.9k
4096B	204.2k	17.3k	13.9k

Threads

Size	inproc	ipc	tcp
64B	353.0k	26.5k	22.5k
256B	341.6k	23.7k	21.3k
1024B	303.8k	24.4k	20.8k
4096B	242.3k	22.8k	20.4k

Latency (req/rep roundtrip)

Async (fibers)

Transport	roundtrips/s	latency
inproc	20.3k	49 µs
ipc	10.0k	100 µs
tcp	9.3k	107 µs

Threads

Transport	roundtrips/s	latency
inproc	8.8k	113 µs
ipc	6.5k	154 µs
tcp	5.9k	168 µs

Notes

• Throughput measures one-way push/pull (no reply needed)
• Latency measures full req/rep roundtrip
• Send/receive uses a nonblocking fast path via zframe_recv_nowait/zframe_send(DONTWAIT), falling back to FD polling when data isn't immediately available
• Async uses Ruby fibers via the async gem
• Threads use a dedicated responder thread
• Async is ~2x faster for inproc latency due to cheap fiber switching

Running

bundle exec ruby --yjit bench/async/throughput.rb
bundle exec ruby --yjit bench/async/latency.rb
bundle exec ruby --yjit bench/threads/throughput.rb
bundle exec ruby --yjit bench/threads/latency.rb