com.github.rholder.moar.concurrent.thread.BalancingThreadPoolExecutor Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of moar-concurrent Show documentation
Show all versions of moar-concurrent Show documentation
This module contains a collection of useful builders and concurrency classes to assist in modeling complex or overly tweakable concurrent processing pipelines.
/*
* Copyright 2012-2013 Ray Holder
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.github.rholder.moar.concurrent.thread;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.AbstractExecutorService;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ThreadPoolExecutor;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
import static java.lang.Math.ceil;
/**
* This is a rough implementation of an auto-balancing thread pool to optimize
* for the following condition: optimal pool size = N * U * (1 + (W/C)) where N
* is the number of CPU's, U is the desired utilization, W is the time each
* thread spends waiting, and C is the time each thread spends using the CPU.
*/
public class BalancingThreadPoolExecutor extends AbstractExecutorService {
private static final int CPUS = Runtime.getRuntime().availableProcessors();
private final float targetUtilization;
private final ConcurrentHashMap liveThreads;
private final ThreadPoolExecutor threadPoolExecutor;
private final ThreadProfiler threadProfiler;
private final AtomicInteger tasksRun;
private final float smoothingWeight;
private final int balanceAfter;
public BalancingThreadPoolExecutor(ThreadPoolExecutor threadPoolExecutor,
ThreadProfiler threadProfiler,
float targetUtilization,
float smoothingWeight,
int balanceAfter) {
if (targetUtilization <= 0.0 || targetUtilization > 1.0) {
throw new IllegalArgumentException();
}
if (threadPoolExecutor == null) {
throw new NullPointerException();
}
this.threadPoolExecutor = threadPoolExecutor;
this.threadProfiler = threadProfiler;
this.targetUtilization = targetUtilization;
this.liveThreads = new ConcurrentHashMap();
this.tasksRun = new AtomicInteger(0);
this.smoothingWeight = smoothingWeight;
this.balanceAfter = balanceAfter;
}
@Override
public void shutdown() {
threadPoolExecutor.shutdown();
}
@Override
public List shutdownNow() {
return threadPoolExecutor.shutdownNow();
}
@Override
public boolean isShutdown() {
return threadPoolExecutor.isShutdown();
}
@Override
public boolean isTerminated() {
return threadPoolExecutor.isTerminated();
}
@Override
public boolean awaitTermination(long timeout, TimeUnit unit) throws InterruptedException {
return threadPoolExecutor.awaitTermination(timeout, unit);
}
@Override
public void execute(final Runnable command) {
// wrap the Runnable such that we can collect profiling on the given tasks
threadPoolExecutor.execute(new Runnable() {
@Override
public void run() {
Thread thisThread = Thread.currentThread();
long threadId = thisThread.getId();
long startTime = threadProfiler.getThreadWaitTime(threadId);
long startCpu = threadProfiler.getThreadCpuTime(threadId);
try {
command.run();
} finally {
Tracking tracking = liveThreads.get(thisThread);
long totalCpuTime = threadProfiler.getThreadCpuTime(threadId) - startCpu;
long totalTime = threadProfiler.getThreadWaitTime(threadId) - startTime;
if(tracking == null) {
// this is an untracked thread, add tracking
tracking = new Tracking();
tracking.avgTotalTime = totalTime;
tracking.avgCpuTime = totalCpuTime;
liveThreads.put(thisThread, tracking);
} else {
// TODO determine exponential smoothing coefficient to specify weight of each task over time
// compute exponential moving averages, see http://en.wikipedia.org/wiki/Exponential_smoothing
tracking.avgTotalTime += smoothingWeight * (totalTime - tracking.avgTotalTime);
tracking.avgCpuTime += smoothingWeight * (totalCpuTime - tracking.avgCpuTime);
}
int count = tasksRun.getAndIncrement();
if(count % balanceAfter == 0) {
balance();
}
}
}
});
}
/**
* Compute and set the optimal number of threads to use in this pool.
*/
private void balance() {
// only try to balance when we're not terminating
if(!isTerminated()) {
Set> threads = liveThreads.entrySet();
long liveAvgTimeTotal = 0;
long liveAvgCpuTotal = 0;
long liveCount = 0;
for (Map.Entry e : threads) {
if (!e.getKey().isAlive()) {
// thread is dead or otherwise hosed
threads.remove(e);
} else {
liveAvgTimeTotal += e.getValue().avgTotalTime;
liveAvgCpuTotal += e.getValue().avgCpuTime;
liveCount++;
}
}
long waitTime = 1;
long cpuTime = 1;
if(liveCount > 0) {
waitTime = liveAvgTimeTotal / liveCount;
cpuTime = liveAvgCpuTotal / liveCount;
}
int size = (int) ceil((CPUS * targetUtilization * (1 + (waitTime / cpuTime))));
size = size > 0 ? size : 1;
size = Math.min(size, threadPoolExecutor.getMaximumPoolSize());
// TODO remove debugging
//System.out.println(waitTime / 1000000 + " ms");
//System.out.println(cpuTime / 1000000 + " ms");
//System.out.println(size);
threadPoolExecutor.setCorePoolSize(size);
}
}
}