Quick benchmark to find out how different channel implementations cope with collecting messages from many tasks.
This compares two scenarios:
- a)
cloned-send: Cloning the sender of a single channel into many tasks - b)
merged-recv: Merging the receivers of many channels withfutures_buffered::FuturesUnordered
I added implementations for flume, tokio::sync::mpsc, and async-channel.
If you find issues with the benchmark, or have suggestions for improvements, please open an issue.
Results on my laptop (Intel 8250u quad-core), in release mode:
running channel benchmark with 1048576 messages and capacity 1024
# multi_thread, tasks: 2, messages per task: 524288, capacity per task: 512
merged-recv tokio 212 ms 202 ns/message
merged-recv flume 233 ms 222 ns/message
merged-recv async-channel 131 ms 125 ns/message
cloned-send tokio 252 ms 240 ns/message
cloned-send flume 200 ms 191 ns/message
cloned-send async-channel 126 ms 121 ns/message
# current_thread, tasks: 2, messages per task: 524288, capacity per task: 512
merged-recv tokio 233 ms 222 ns/message
merged-recv flume 120 ms 115 ns/message
merged-recv async-channel 163 ms 155 ns/message
cloned-send tokio 162 ms 154 ns/message
cloned-send flume 73 ms 70 ns/message
cloned-send async-channel 118 ms 112 ns/message
# multi_thread, tasks: 16, messages per task: 65536, capacity per task: 64
merged-recv tokio 359 ms 342 ns/message
merged-recv flume 443 ms 423 ns/message
merged-recv async-channel 417 ms 397 ns/message
cloned-send tokio 506 ms 482 ns/message
cloned-send flume 834 ms 795 ns/message
cloned-send async-channel 529 ms 505 ns/message
# current_thread, tasks: 16, messages per task: 65536, capacity per task: 64
merged-recv tokio 282 ms 269 ns/message
merged-recv flume 133 ms 127 ns/message
merged-recv async-channel 170 ms 162 ns/message
cloned-send tokio 183 ms 174 ns/message
cloned-send flume 80 ms 77 ns/message
cloned-send async-channel 157 ms 150 ns/message
# multi_thread, tasks: 64, messages per task: 16384, capacity per task: 16
merged-recv tokio 441 ms 421 ns/message
merged-recv flume 418 ms 398 ns/message
merged-recv async-channel 393 ms 374 ns/message
cloned-send tokio 886 ms 845 ns/message
cloned-send flume 3590 ms 3423 ns/message
cloned-send async-channel 1890 ms 1803 ns/message
# current_thread, tasks: 64, messages per task: 16384, capacity per task: 16
merged-recv tokio 339 ms 323 ns/message
merged-recv flume 149 ms 142 ns/message
merged-recv async-channel 162 ms 154 ns/message
cloned-send tokio 218 ms 208 ns/message
cloned-send flume 92 ms 88 ns/message
cloned-send async-channel 392 ms 374 ns/message
# multi_thread, tasks: 256, messages per task: 4096, capacity per task: 4
merged-recv tokio 438 ms 418 ns/message
merged-recv flume 395 ms 377 ns/message
merged-recv async-channel 357 ms 341 ns/message
cloned-send tokio 914 ms 871 ns/message
cloned-send flume 7121 ms 6791 ns/message
cloned-send async-channel 2181 ms 2080 ns/message
# current_thread, tasks: 256, messages per task: 4096, capacity per task: 4
merged-recv tokio 461 ms 440 ns/message
merged-recv flume 189 ms 180 ns/message
merged-recv async-channel 162 ms 154 ns/message
cloned-send tokio 389 ms 371 ns/message
cloned-send flume 162 ms 155 ns/message
cloned-send async-channel 511 ms 487 ns/message
# multi_thread, tasks: 512, messages per task: 2048, capacity per task: 2
merged-recv tokio 432 ms 412 ns/message
merged-recv flume 398 ms 380 ns/message
merged-recv async-channel 259 ms 247 ns/message
cloned-send tokio 927 ms 884 ns/message
cloned-send flume 11345 ms 10820 ns/message
cloned-send async-channel 2151 ms 2052 ns/message
# current_thread, tasks: 512, messages per task: 2048, capacity per task: 2
merged-recv tokio 446 ms 425 ns/message
merged-recv flume 212 ms 203 ns/message
merged-recv async-channel 139 ms 132 ns/message
cloned-send tokio 358 ms 341 ns/message
cloned-send flume 319 ms 304 ns/message
cloned-send async-channel 621 ms 592 ns/message