msgpack[C++]使用笔记 和 msgpack/cPickle性能对比

msgpack[C++]使用笔记 和 msgpack/cPickle性能对比

转载自： http://blog.csdn.net/xiarendeniao/article/details/6801338

python版本msgpack安装：

wget http://pypi.python.org/packages/source/m/msgpack-python/msgpack-python-0.1.9.tar.gz

python2.x setup.py install --prefix=/usr/local/similarlib/

python版本的msgpack灰常好用，速度上比python内置的pickle和cpickle都要快一些，C++版本的使用比较麻烦，下面是本人学习时的一个demo，解析python-msgpack dump的一个复杂字典。

#include <msgpack.hpp>

#include <fstream>
#include <iostream>
using namespace std;

template <class T>
void msgunpack(const char* binary_file, T& t, char* buff, uint32_t max){
    msgpack::unpacked msg;
    ifstream tf_file(binary_file,ios::in|ios::binary|ios::ate);
    uint32_t size = tf_file.tellg();
    tf_file.seekg(0, ios::beg);
    tf_file.read(buff, size);
    tf_file.close();
    msgpack::unpack(&msg, buff, size);
    msg.get().convert(&t);
}


typedef map<uint32_t, uint32_t> WordsMap;
typedef map<uint32_t, WordsMap> FieldsMap;
typedef map<uint64_t, FieldsMap> DocsMap;

int main(int argc, char** argv)
{
    uint32_t MAX_BUFF = 1024*1024*100; //100MB
    char* BUFF = new char[MAX_BUFF];

    DocsMap docsMap;
    msgpack::unpacked msg;
    msgunpack("/data/wikidoc/tf_dict_for_nodes/1-1000", docsMap, BUFF, MAX_BUFF);
    //        msg.get().convert(&docsMap);
    cout << docsMap.size() << endl;
        delete[] BUFF;
}

参考： http://wiki.msgpack.org/pages/viewpage.action?pageId=1081387#QuickStartforC%2B%2B-ImplementationStatus

下面是本人自己封装的一个msgpack接口头文件mymsgpack.h

 #ifndef MY_MSGPACK_H

#ifndef MY_MSGPACK_H
#define MY_MSGPACK_H
#include <fstream>
#include <msgpack.hpp>
using namespace std;

template <class T>
void load_from_file(const char* binary_file, T& t) {
        ifstream binaryFstream(binary_file,ios::in|ios::binary|ios::ate);
        uint32_t size = binaryFstream.tellg();
        char* buff = new char[size];
        binaryFstream.seekg(0, ios::beg);
        binaryFstream.read(buff, size);
        binaryFstream.close();
        msgpack::unpacked msg;
        msgpack::unpack(&msg, buff, size);
        msg.get().convert(&t);
        delete[] buff;
}

template <class T>
void load_from_str(const char* binary_str, int len, T& t) {
        msgpack::unpacked msg;
        msgpack::unpack(&msg, binary_str, len);
        msg.get().convert(&t);
}

template <class T>
void dump_to_file(T& t, const char* dump_file) {
    msgpack::sbuffer sbuf;
    msgpack::pack(sbuf, t);
    ofstream dumpFstream(dump_file, ios::out|ios::binary|ios::trunc);
    dumpFstream.write(sbuf.data(), sbuf.size());
    dumpFstream.close();
}

template <class T>
void dump_to_str(T& t, char** dump_str, int& len) { //外部释放*dump_str
    msgpack::sbuffer sbuf;
    msgpack::pack(sbuf, t);
    len = sbuf.size();
    *dump_str = (char*)malloc(sbuf.size());
    memcpy(*dump_str, sbuf.data(), sbuf.size());
}

#endif

 

msgpack编译通过，链接不上的问题 undefined reference to `__sync_sub_and_fetch_4'

在x84_64机器上正常，在32bit机器上出现上述问题

[xudongsong@BigServerU-4 msgpack-0.5.7]$ cat /etc/issue
CentOS release 5.4 (Final)
Kernel \r on an \m

[xudongsong@BigServerU-4 msgpack-0.5.7]$ file /sbin/init
/sbin/init: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), for GNU/Linux 2.6.9, dynamically linked (uses shared libs), for GNU/Linux 2.6.9, stripped

./configure不报错，但是查看config.log显示有错误，程序链接msgpack的库时也报错

原因：gcc不能识别CPU体系，需要手动指明

[xudongsong@BigServerU-4 msgpack-0.5.7]$ CFLAGS="-march=pentium -mtune=pentium" ./configure --prefix=/home/xudongsong/msgpack_static --enable-static=yes --enable-shared=no

make, make install

[xudongsong@BigServerU-4 jobs]$ g++ job_calc_weight.cpp -o job_calc_weight -I/home/xudongsong/msgpack_static/include/ -L/home/xudongsong/msgpack_static/lib/ -lmsgpack

通过！

 

 

下面是msgpack和cPickle进行性能pk的demo程序（不比较pickle，是因为它比cPickle更慢，《Python cook book》里面有说明）：

mport sys,time,msgpack,pickle,cPickle,random

test_list = []
i = 0
while i<100000:
    test_list = random.randrange(1,100000)
    i += 1

print "common len(serialize) = %s"%len(cPickle.dumps(test_list,0))
print "compress len(serialize) = %s"%len(cPickle.dumps(test_list,1))

#------------------------------------------------------------------------
results = {}
time_start = time.time()
for i in range(1,1000000):
        results[i] = cPickle.dumps(test_list,1)
time_mid_1 = time.time()
print "cPickle dumps eats %s s"%str(time_mid_1-time_start)

for i in range(1,1000000):
    cPickle.loads(results[i])
time_mid_2 = time.time()
print "cPickle loads eats %s s"%str(time_mid_2-time_mid_1)

#------------------------------------------------------------------------
results = {}
time_start = time.time()
for i in range(1,1000000):
    results[i] = msgpack.dumps(test_list)
time_mid_1 = time.time()
print "msgpack pack eats %s s"%str(time_mid_1-time_start)

for i in range(1,1000000):
    msgpack.loads(results[i])
time_mid_2 = time.time()
print "msgpack unpack eats %s s"%str(time_mid_2-time_mid_1)