Skip to content

bench

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

bench

README.md

Benchmarks

Measured on Linux x86_64, Ruby 4.0.2 +YJIT. Each cell is the fastest of 3 timed rounds (~1 s each) after a calibration warmup, so transient scheduler/GC jitter is filtered out. Between-run variance on the same machine is ~5-15 % depending on transport; treat single-digit deltas across runs as noise.

Regenerate the tables below from the latest run in results.jsonl:

ruby bench/report.rb --update-readme

Throughput (PUSH/PULL, msg/s)

┌──────┐       ┌──────┐
│ PUSH │──────→│ PULL │
└──────┘       └──────┘

1 peer

Message size inproc ipc tcp
64 B 625.1k msg/s / 40.0 MB/s 187.1k msg/s / 12.0 MB/s 235.3k msg/s / 15.1 MB/s
1 KiB 612.5k msg/s / 627 MB/s 180.1k msg/s / 184 MB/s 179.9k msg/s / 184 MB/s
8 KiB 658.4k msg/s / 5.39 GB/s 91.5k msg/s / 750 MB/s 94.9k msg/s / 777 MB/s
64 KiB 652.9k msg/s / 42.79 GB/s 20.3k msg/s / 1.33 GB/s 17.8k msg/s / 1.17 GB/s

3 peers

Message size inproc ipc tcp
64 B 487.4k msg/s / 31.2 MB/s 123.1k msg/s / 7.88 MB/s 200.9k msg/s / 12.8 MB/s
1 KiB 491.6k msg/s / 503 MB/s 140.1k msg/s / 144 MB/s 164.2k msg/s / 168 MB/s
8 KiB 490.0k msg/s / 4.01 GB/s 84.0k msg/s / 688 MB/s 85.5k msg/s / 700 MB/s
64 KiB 478.1k msg/s / 31.33 GB/s 18.7k msg/s / 1.22 GB/s 17.7k msg/s / 1.16 GB/s

Round-trip latency (REQ/REP, µs)

┌─────┐  req   ┌─────┐
│ REQ │───────→│ REP │
│     │←───────│     │
└─────┘  rep   └─────┘

Round-trip = one req.send + one req.receive + matching rep ops. Latency is 1 / msgs_s converted to µs.

Message size inproc ipc tcp
64 B 15.0 µs 58.7 µs 72.4 µs
1 KiB 14.9 µs 63.3 µs 77.0 µs
8 KiB 15.3 µs 70.2 µs 82.2 µs
64 KiB 14.9 µs 108 µs 133 µs

io_uring

With liburing-dev installed, io-event uses io_uring instead of epoll. Inproc throughput jumps significantly. IPC and TCP are within variance.

# Debian/Ubuntu
sudo apt install liburing-dev
gem pristine io-event

Running

# Full suite (one run_id shared across patterns for cross-pattern comparison)
RUN_ID=$(date +%Y-%m-%dT%H:%M:%S)
for d in push_pull req_rep router_dealer pub_sub; do
  OMQ_BENCH_RUN_ID=$RUN_ID bundle exec ruby --yjit bench/$d/omq.rb
done

# Regression report (latest vs previous run)
bundle exec ruby bench/report.rb

# Regenerate README tables from the latest run
bundle exec ruby bench/report.rb --update-readme

# Full comparison table
bundle exec ruby bench/report.rb --all