Thread: Multithreading File Reading

I was reading up on multithreading and I came across things like Mutexes, Locks, Starvation and it's quite interesting. Here I'm using CRITICAL_SECTION in my code to prevent data corruption and so forth. I could have used Mutexex, Events, Semaphores, etc... I managed to read and add the contents of a 5MB text file in about 5.9 seconds with 7 threads. This is a big improvement since I had a program before that took a whole 10 seconds to read and add the contents of a 1MB file to a vector. This one uses maps and counts each word's/phrase occurrences. The speed is highly dependent on the placement of the critical sections. I shaved off about 4 seconds just by moving the EnterCriticalSection down one line!!

I'm looking for other ways to optimize the CRITICAL_SECTIONS though. I want to get this down to at least 3 seconds if possible. xD

Code:

#include <windows.h>
#include <process.h>
#include <iostream>
#include <string>
#include <fstream>
#include <map>
#include <time.h>

using namespace std;

map <string, long> words;
ifstream f;
CRITICAL_SECTION c, c2;

unsigned int __stdcall thread(void *)
{
    string buffer;
    map <string, long>::iterator iter;
    while (f >> buffer) {
        EnterCriticalSection(&c);
        f >> buffer;
        LeaveCriticalSection(&c);
        iter = words.find(buffer);
        EnterCriticalSection(&c2);
        if (iter != words.end())
            iter->second++;
        else
            words.insert(make_pair(buffer, 1));
        LeaveCriticalSection(&c2);

    }
    return 0;
}

int main()
{
    cout << "How much threads do you want to use? -> ";
    int thread_count;
    cin >> thread_count;
    f.open("file.txt", ios::in);
    clock_t t1, t2;
    if (f.is_open())
    {
        InitializeCriticalSection(&c);
        InitializeCriticalSection(&c2);
        t1 = clock();
        HANDLE * hThreads = (HANDLE *)malloc(sizeof(HANDLE) * thread_count);
        for(int i = 0; i < thread_count; i++)
            hThreads[i] = (HANDLE) _beginthreadex(0, 0, &thread, (void *)0, 0, 0);
        WaitForMultipleObjects(thread_count, hThreads, true, INFINITE);
        t2 = clock();
        for(int i = 0; i < thread_count; i++)
	     CloseHandle(hThreads[i]);
        DeleteCriticalSection(&c2);
        DeleteCriticalSection(&c);
        f.close();
        map<string, long>::iterator iter = words.begin();
        //IF YOU WISH TO DISPLAY THE MAP
        /*for(; iter != words.end(); iter++)
            cout << iter->first.c_str() << " - " << iter->second << endl;*/
        cout << "Unique WOrD CoUNt -> " << words.size() << endl;
    }
    else
    {
        cout << "The file couldn't be opened" << endl;
    }
    cout << "Word count took " << (double(t2 - t1) / 1000) 
           << " second(s) with " << thread_count << " threads";
    words.clear();
    std::cin.ignore();
    std::cin.get();
    return 0;
}

just about 9 threads you'll get the best performance and oh yea please change the file name before running it