c++ - "std::map with mutexes"与 "libcds maps (Michael Hashmap and Split Order List)"并行插入、查找、删除之间是否有任何速度测试？

Question

所以我真的很想看到一些并行的速度测试(比如从 100 到 10000 个并行线程)，其中每个线程至少在 3 种类型的并发映射上插入、查找、删除 - std::map(有一些互斥锁)与 libcds (Concurrent Data Structures) ...

例如，如果这样的比较尚不存在，请帮助我创建一个。

直接相关: LibCds: Michael Hashmap and Split Order List

假设我们有

#include <iostream>
#include <boost/thread.hpp>
#include <map>

class TestDs 
{
public:

    virtual bool containsKey(int key)=0;
    virtual int get(int key)=0;
    virtual int put(int key, int value)=0;
    virtual int remove(int key)=0;

    virtual int size()=0;
    virtual const char* name()=0;
    virtual void print()=0;
    virtual void shutdown()=0;
};

class GeneralMap: public TestDs
{
private:

    std::map<int,int> _ds;
    mutable boost::mutex mut_;
public:
    GeneralMap() {}

    bool containsKey(int key) {
        boost::mutex::scoped_lock lock(mut_);
        if ( _ds.find(key) != _ds.end())
        {
            return true;
        }
        else
        {
            return false;
        }
    }

    int get(int key) {
        boost::mutex::scoped_lock lock(mut_);
        return _ds[key];
    }

    int put(int key, int value) {
        boost::mutex::scoped_lock lock(mut_);
        _ds.insert(std::pair<int, int>(key,value));
        return key;
    }

    int remove(int key) {
        boost::mutex::scoped_lock lock(mut_);
        return _ds.erase(key);
    }

    int size() {
        boost::mutex::scoped_lock lock(mut_);
        return _ds.size();
    }
    const char* name() {
        return "StdMap";
    }
    void print() {}
    void shutdown() {}

};

比起如何创建这样的测试，它会创建 N 个线程，每个线程都会调用创建、查找删除...我开始写一些东西，但它现在可以用 boost 1.47.0 编译代码...

#include <iostream>
#include <boost/thread.hpp>
#include <map>
#include <boost/thread.hpp>
#include <boost/thread/locks.hpp>
#include <boost/date_time.hpp>
#include <boost/random/mersenne_twister.hpp>
#include <boost/random/uniform_int_distribution.hpp>
#include <boost/random.hpp>
#include <boost/progress.hpp>

class timer 
{ 
public: 
    timer() : start_time_(boost::posix_time::microsec_clock::local_time()) {} 

    void restart() 
    {
        start_time_ = boost::posix_time::microsec_clock::local_time();
    } 

    boost::posix_time::time_duration elapsed() const 
    {
        return boost::posix_time::microsec_clock::local_time() - start_time_;
    } 
private: 
    boost::posix_time::ptime start_time_; 
};

class TestDs 
{
public:

    virtual bool containsKey(int key)=0;
    virtual int get(int key)=0;
    virtual int put(int key, int value)=0;
    virtual int remove(int key)=0;

    virtual int size()=0;
    virtual const char* name()=0;
    virtual void print()=0;
    virtual void shutdown()=0;
};

class GeneralMap: public TestDs
{
private:

    std::map<int,int> _ds;
    mutable boost::mutex mut_;
public:
    GeneralMap() {}

    bool containsKey(int key) {
        boost::mutex::scoped_lock lock(mut_);
        if ( _ds.find(key) != _ds.end())
        {
            return true;
        }
        else
        {
            return false;
        }
    }

    int get(int key) {
        boost::mutex::scoped_lock lock(mut_);
        return _ds[key];
    }

    int put(int key, int value) {
        boost::mutex::scoped_lock lock(mut_);
        _ds.insert(std::pair<int, int>(key,value));
        return key;
    }

    int remove(int key) {
        boost::mutex::scoped_lock lock(mut_);
        return _ds.erase(key);
    }

    int size() {
        boost::mutex::scoped_lock lock(mut_);
        return _ds.size();
    }
    const char* name() {
        return "StdMap";
    }
    void print() {}
    void shutdown() {}

};

template <class map_wraper_t>
class test_map_wraper
{
public:

    test_map_wraper(int threads_number)
    {
        n = threads_number;
    }

    void start_tests()
    {

        boost::upgrade_lock<boost::shared_mutex> lock(tests);
        boost::upgrade_to_unique_lock<boost::shared_mutex> uniqueLock(lock);

        boost::shared_lock<boost::shared_mutex> lock_r(results);

        for(int i=0; i<n; i++)
        {
            boost::thread worker(&test_map_wraper::test, this, i);
        }
        boost::thread worker_r(&test_map_wraper::result, this);
        timerForCaptureFame.restart();
    }
private:
    int n;
    boost::shared_mutex  tests;
    boost::shared_mutex  results;
    boost::random::mt19937 rng;
    timer timerForCaptureFame;
    map_wraper_t Ds;
    boost::progress_display *show_progress;

    void test( int i)
    {
        boost::shared_lock<boost::shared_mutex> lock_r(results);
        boost::shared_lock<boost::shared_mutex> lock(tests);
        Ds.put(i, 0);
        if (Ds.containsKey(i))
        {
            Ds.get(i);
        }
        Ds.remove(i);
    }

    void result()
    {
        boost::upgrade_lock<boost::shared_mutex> lock(results);
        boost::upgrade_to_unique_lock<boost::shared_mutex> uniqueLock(lock);

        std::cout <<  std::endl << "test of " << Ds.name() << " complite;" << std::endl << "test performed on " << n << " items" << std::endl << "test duration: " << timerForCaptureFame.elapsed() << std::endl;
    }
};


int main()
{
    int threads_n = 1000;
    int tests = 5;
    std::cout << "Number of required tests: " << tests << std::endl << "Number of threads in each test: " << threads_n << std::endl << "Wait for it..." << std::endl;
    //for(int i = 0; i < tests; ++i)
    //{
        test_map_wraper<GeneralMap> GeneralMapTest(threads_n);
        GeneralMapTest.start_tests();
    //}
    std::cin.get();
    return 0;
}

Answer 1

是的，它在 LibCds 的单元测试中

它将使用各种不同的 map 类型运行相同的测试场景，包括您的无锁实现，还有带锁和不带锁的 std::map/set。

当然，没有同步显然是赢家，但是当混合中有编写器时它就不起作用了。无锁实现比具有同步访问的 STL 容器快得多。

所有这些都不足为奇。你为什么要问？

附言。单元测试在这里:

• 获取 tar 球
• 摘录
• 构建(cd build && ./build.sh)
• 在 ./bin/*/cds-unit 或 ./bin/*/cds-unit-debug 中进行单元测试(如果使用 --debug 构建) -测试