package com.coralblocks.coralqueue.bench; import com.coralblocks.coralqueue.AtomicQueue; import com.coralblocks.coralqueue.Queue; import com.coralblocks.coralqueue.util.MutableLong; import com.coralblocks.coralqueue.util.PauseSupport; import com.coralblocks.coralqueue.waitstrategy.SpinWaitStrategy; import com.coralblocks.coralqueue.waitstrategy.WaitStrategy; import com.coralblocks.coralthreads.Affinity; import com.coralblocks.coralutils.SystemUtils; import com.coralblocks.coralutils.bench.Benchmarker; import com.coralblocks.coralutils.tsutils.Timestamper; /** * Different cores: (without hyper-threading) * * java -server -verbose:gc -cp target/coralqueue-all.jar -Xms2g -Xmx8g -XX:NewSize=512m -XX:MaxNewSize=1024m -DmessagesToWarmup=1000000 -DmessagesToTest=10000000 -Dpause=100 -DproducerProcToBind=2 -DconsumerProcToBind=3 -DblockCount=true -DdetailedBenchmarker=true -DexcludeNanoTimeCost=true com.coralblocks.coralqueue.bench.TestProducerLatency * * Iterations: 10000000 | Avg Time: 24.65 nanos | Min Time: 4.0 nanos | Max Time: 6.03 micros | 75% = [avg: 24.0 nanos, max: 25.0 nanos] | 90% = [avg: 24.0 nanos, max: 25.0 nanos] | 99% = [avg: 24.0 nanos, max: 26.0 nanos] | 99.9% = [avg: 24.0 nanos, max: 148.0 nanos] | 99.99% = [avg: 24.0 nanos, max: 264.0 nanos] | 99.999% = [avg: 24.0 nanos, max: 415.0 nanos] * * Same core: (with hyper-threading) * * java -server -verbose:gc -cp target/coralqueue-all.jar -Xms2g -Xmx8g -XX:NewSize=512m -XX:MaxNewSize=1024m -DmessagesToWarmup=1000000 -DmessagesToTest=10000000 -Dpause=100 -DproducerProcToBind=2 -DconsumerProcToBind=6 -DblockCount=true -DdetailedBenchmarker=true -DexcludeNanoTimeCost=true com.coralblocks.coralqueue.bench.TestProducerLatency * * Iterations: 10000000 | Avg Time: 13.91 nanos | Min Time: 7.0 nanos | Max Time: 10.248 micros | 75% = [avg: 13.0 nanos, max: 15.0 nanos] | 90% = [avg: 13.0 nanos, max: 16.0 nanos] | 99% = [avg: 13.0 nanos, max: 27.0 nanos] | 99.9% = [avg: 13.0 nanos, max: 32.0 nanos] | 99.99% = [avg: 13.0 nanos, max: 281.0 nanos] | 99.999% = [avg: 13.0 nanos, max: 1.031 micros] */ public class TestProducerLatency { public static void main(String[] args) throws InterruptedException { final int messagesToWarmup = SystemUtils.getInt("messagesToWarmup", 1000000); final int messagesToTest = SystemUtils.getInt("messagesToTest", 10000000); final long pause = SystemUtils.getLong("pause", 100); int prodProcToBind = SystemUtils.getInt("producerProcToBind", 2); int consProcToBind = SystemUtils.getInt("consumerProcToBind", 3); /* * messagesToWarmup is VERY important to warmup benchmark class. Failure to pass it will producer outliers !!! */ final Benchmarker bench = Benchmarker.create(messagesToWarmup); final Timestamper timestamper = bench.getTimestamper(); final Queue<MutableLong> queue = new AtomicQueue<MutableLong>(512, MutableLong.BUILDER); // we use a strategy to check block count final WaitStrategy producerWaitStrategy = new SpinWaitStrategy(); System.out.println("messagesToWarnup: " + messagesToWarmup); System.out.println("messagesToTest: " + messagesToTest); System.out.println("pause: " + pause); System.out.println("timestamper: " + timestamper); Thread producer = new Thread(new Runnable() { @Override public void run() { Affinity.bind(); long count = 0; long total = messagesToWarmup + messagesToTest; while(count < total) { MutableLong ml = null; bench.mark(); while((ml = queue.nextToDispatch()) == null) { producerWaitStrategy.block(); } ml.set(count); queue.flush(true); // lazySet true bench.measure(); producerWaitStrategy.reset(); PauseSupport.assignment(pause); count++; } Affinity.unbind(); System.out.println("producer exiting..."); } }, "Producer"); Thread consumer = new Thread(new Runnable() { @Override public void run() { Affinity.bind(); long total = messagesToWarmup + messagesToTest; boolean running = true; while (running) { long avail = queue.availableToPoll(); if (avail > 0) { for(int i = 0; i < avail; i++) { MutableLong ml = queue.poll(); long x = ml.get(); if (x == total - 1) running = false; } queue.donePolling(false); // no lazy, tell producer immediately in case queue was full... } } Affinity.unbind(); System.out.println("consumer exiting..."); } }, "Consumer"); if (Affinity.isAvailable()) { Affinity.assignToProcessor(prodProcToBind, producer); Affinity.assignToProcessor(consProcToBind, consumer); } else { System.err.println("Thread affinity not available!"); } consumer.start(); producer.start(); consumer.join(); producer.join(); System.out.println("Producer block count: " + producerWaitStrategy.getTotalBlockCount()); System.out.println(bench.results()); } }