Compare concurrency control in Coroutine VS Reactive VS VirtualThread(a.k.a Loom)

compare_concurrency_control.kt

package org.example

import kotlinx.coroutines.*
import kotlinx.coroutines.channels.Channel
import kotlinx.coroutines.flow.*
import kotlinx.coroutines.sync.Semaphore
import kotlinx.coroutines.sync.withPermit
import org.openjdk.jmh.annotations.*
import java.util.concurrent.TimeUnit

fun main(args: Array<String>) {
    org.openjdk.jmh.Main.main(args);
}

@Warmup(iterations = 7, time = 1)
@Measurement(iterations = 5, time = 1)
@BenchmarkMode(Mode.All)
@OutputTimeUnit(TimeUnit.MILLISECONDS)
@State(Scope.Benchmark)
@Fork(1)
open class Compare {
    val scope = GlobalScope
    val channel = Channel<CompletableDeferred<Unit>>(128)
    init {
        scope.launch {
            channel
                .receiveAsFlow()
                .cancellable()
                .flatMapMerge(16) {
                    flow<Nothing> {
                        delay(0)
                        it.complete(Unit)
                    }
                }
                .collect()
        }
    }

    @Benchmark
    fun reactive() = runBlocking {
        val deferred = CompletableDeferred<Unit>()
        channel.send(deferred)
        deferred.join()
    }

    val wait = Semaphore(128)
    val concurrency = Semaphore(16)

    @Benchmark
    fun kotlinSemaphore() = runBlocking {
        wait.withPermit {
            concurrency.withPermit {
                delay(0)
            }
        }
    }

    val jwait = java.util.concurrent.Semaphore(128)
    val jconcurrency = java.util.concurrent.Semaphore(16)
    @Benchmark
    fun loomSemaphore() = Thread.ofVirtual().start {
        jwait.acquire()
        jconcurrency.acquire()
        Thread.sleep(0)
        jconcurrency.release()
        jwait.release()
    }.join()
}

jmh-results.txt

Benchmark                          Mode      Cnt     Score     Error   Units
Compare.kotlinSemaphore           thrpt        5  5896.626 ± 195.969  ops/ms
Compare.loomSemaphore             thrpt        5   326.712 ±  11.815  ops/ms
Compare.reactive                  thrpt        5   296.230 ±  44.611  ops/ms
Compare.kotlinSemaphore            avgt        5     0.002 ±   0.001   ms/op
Compare.loomSemaphore              avgt        5     0.040 ±   0.022   ms/op
Compare.reactive                   avgt        5     0.041 ±   0.011   ms/op
Compare.kotlinSemaphore          sample  1949362     0.003 ±   0.001   ms/op
Compare.kotlinSemaphore:p0.00    sample             ≈ 10⁻⁴             ms/op
Compare.kotlinSemaphore:p0.50    sample              0.002             ms/op
Compare.kotlinSemaphore:p0.90    sample              0.002             ms/op
Compare.kotlinSemaphore:p0.95    sample              0.003             ms/op
Compare.kotlinSemaphore:p0.99    sample              0.003             ms/op
Compare.kotlinSemaphore:p0.999   sample              0.084             ms/op
Compare.kotlinSemaphore:p0.9999  sample              1.426             ms/op
Compare.kotlinSemaphore:p1.00    sample             16.908             ms/op
Compare.loomSemaphore            sample   825468     0.037 ±   0.001   ms/op
Compare.loomSemaphore:p0.00      sample              0.001             ms/op
Compare.loomSemaphore:p0.50      sample              0.032             ms/op
Compare.loomSemaphore:p0.90      sample              0.056             ms/op
Compare.loomSemaphore:p0.95      sample              0.070             ms/op
Compare.loomSemaphore:p0.99      sample              0.147             ms/op
Compare.loomSemaphore:p0.999     sample              0.485             ms/op
Compare.loomSemaphore:p0.9999    sample              2.693             ms/op
Compare.loomSemaphore:p1.00      sample             42.992             ms/op
Compare.reactive                 sample   723234     0.041 ±   0.001   ms/op
Compare.reactive:p0.00           sample              0.001             ms/op
Compare.reactive:p0.50           sample              0.037             ms/op
Compare.reactive:p0.90           sample              0.069             ms/op
Compare.reactive:p0.95           sample              0.082             ms/op
Compare.reactive:p0.99           sample              0.127             ms/op
Compare.reactive:p0.999          sample              0.213             ms/op
Compare.reactive:p0.9999         sample              1.148             ms/op
Compare.reactive:p1.00           sample              3.375             ms/op
Compare.kotlinSemaphore              ss        5     0.023 ±   0.008   ms/op
Compare.loomSemaphore                ss        5     0.068 ±   0.098   ms/op
Compare.reactive                     ss        5     0.252 ±   0.156   ms/op

$ sysctl machdep.cpu
machdep.cpu.cores_per_package: 12
machdep.cpu.core_count: 12
machdep.cpu.logical_per_package: 12
machdep.cpu.thread_count: 12
machdep.cpu.brand_string: Apple M2 Max