使用环形缓冲提升IO效率(无锁缓冲)

摘自:http://www.oschina.net/code/snippet_54334_12505

代码源于 http://www.ibm.com/developerworks/cn/linux/l-cn-lockfree/ 的实现. 
注意: 构造时参数 buf_size 必须是2的N次方.

 

#ifndef min
#define min(a,b) (((a) < (b)) ? (a) : (b))
#endif

#ifndef max
#define max(a,b) (((a) > (b)) ? (a) : (b))
#endif

// 无锁缓冲队列.
class circular_buffer
{
public:
   circular_buffer(int buf_size)
      : m_buffer_size(buf_size)
      , m_circle_buffer(NULL)
      , m_write_p(0)
      , m_read_p(0)
   {
      m_circle_buffer = new char[m_buffer_size];
   }

   ~circular_buffer()
   {
      if (m_circle_buffer)
         delete[] m_circle_buffer;
      m_circle_buffer = NULL;
   }

   void clear()
   {
      m_write_p = 0;
      m_read_p = 0;
   }

   unsigned int available()
   {
      return m_buffer_size - (m_write_p - m_read_p);
   }

   unsigned int used()
   {
      return m_write_p - m_read_p;
   }

   unsigned int put_data(char* buffer, unsigned int len)
   {
      unsigned int l;
      len = _min(len, m_buffer_size - m_write_p + m_read_p);
      /* first put the data starting from fifo->in to buffer end */
      l = _min(len, m_buffer_size - (m_write_p & (m_buffer_size - 1)));
      memcpy(m_circle_buffer + (m_write_p & (m_buffer_size - 1)), buffer, l);
      /* then put the rest (if any) at the beginning of the buffer */
      memcpy(m_circle_buffer, buffer + l, len - l);
      m_write_p += len;
      return len;
   }

   unsigned int get_data(char* buffer, unsigned int len)
   {
      unsigned int l; 
      len = _min(len, m_write_p - m_read_p); 
      /* first get the data from fifo->out until the end of the buffer */ 
      l = _min(len, m_buffer_size - (m_read_p & (m_buffer_size - 1))); 
      memcpy(buffer, m_circle_buffer + (m_read_p & (m_buffer_size - 1)), l); 
      /* then get the rest (if any) from the beginning of the buffer */ 
      memcpy(buffer + l, m_circle_buffer, len - l); 
      m_read_p += len; 
      return len; 
   }

protected:
   inline unsigned int _max(unsigned int a, unsigned int b)
   {
      return max(a, b);
   }

   inline unsigned int _min(unsigned int a, unsigned int b)
   {
      return min(a, b);
   }

private:
   int m_buffer_size;
   char* m_circle_buffer;
   unsigned int m_write_p;
   unsigned int m_read_p;
};

 

 

摘自:http://blog.csdn.net/xocoder/article/details/7880769

 

最近在重构之前写的网络底层时,从各个方面认真考虑了每一个细节实现。其中,在提交I/O(WSASend/WSARecv)和I/O完成(GetQueuedCompletionStatus)时,难免出现一个缓冲区需要两个线程公用的问题。

 

假设主线程不断发送该消息,这些消息被堆叠在一个缓冲区里,定时使用WSASend提交发送I/O请求,在GetQueuedCompletionStatus返回后,才能按照已发送的字节数去删掉该缓冲区里相应字节数的数据。不明白?好吧我说的简单一些。

WSASend调用后,你传递的参数只是说明:我希望发送这么多数据。但请求提交后,你的要求未必能够被全部满足,也就是说也许你想发送1024字节的东西,但也许GetQueuedCompletionStatus返回,操作完成后,本次只成功发送了1000个字节,也就是说剩余的24字节的数据,你还需要再调用WSASend,直到都发送成功为止。所以在这种情况下,一定要等待GetQueuedCompletionStatus返回,才知道究竟发送成功了多少,也才能从之前的发送缓冲里删掉数据。否则,如果你在提交WSASend时就把数据删掉了,而GetQueuedCompletionStatus返回后却告诉你只发了1000字节,那就杯具了-----那24字节的数据永远地离开了我们。

 

而在这种情况下,我们发送消息,也就是向这个缓冲区后面堆放要发送的数据,是主线程中执行的,而GetQueuedCompletionStatus完成后,从缓冲区内弹出数据,确是IOCP的工作线程做的。当然,最简单的办法就是------加个锁呗。但是,在I/O频繁的情况下,可以想象会出现多少线程争用的情况。于是,就有了本文要说的东西:环形缓冲。

 

环形缓冲的原理并不难理解,只适用于一个线程写,一个线程读的情况。环形缓冲的原理我就不再赘述,可以自行搜索。

 

废话不多说。下面给出我在这次优化中写的一个环形缓冲类,该环形缓冲完美地在IOCP中工作了起来,实实在在地解决了线程争用引发地效率低下。

2012.9.1 0:57 重贴代码
修改了一个可能出现的误置Full标志的BUG,之前的代码中,是先增加写指针,再判断是否等于读指针,等于则置Full标志,但若在该判断之前,读线程将数据读空,此时写线程继续工作,进行该判断时,就会发现写指针 = 读指针(但是是由于读空造成的),于是错误地将状态置为Full。

#pragma once

#include "XBaseDefine.h"

enum eXRingBufferState
{
	eXRingBufferState_None = 0,
	eXRingBufferState_Empty = 1,
	eXRingBufferState_Full = 2,
	eXRingBufferState_Max
};

class XRingBuffer
{
public:
	XRingBuffer(const DWORD size);
	~XRingBuffer();
	
	bool pushData(const char* data, const DWORD size);
	bool copyData(char* dest, const DWORD destSize, const DWORD size);
	bool popData(char* dest, const DWORD destSize, const DWORD popSize);
	bool popData(const DWORD popSize);

	const DWORD getUsedSize() const;
	const DWORD getFreeSize() const;
private:
	DWORD _size;
	char* _buffer;
	volatile DWORD _write_pos;
	volatile DWORD _read_pos;
	volatile DWORD _state;
};

 

 

#include "XRingBuffer.h"

XRingBuffer::XRingBuffer(const DWORD size) :
_size(size),
_buffer(NULL),
_write_pos(0),
_read_pos(0),
_state(eXRingBufferState_Empty)
{
	_buffer = new char[size];
}

XRingBuffer::~XRingBuffer()
{
	delete [] _buffer;
	_buffer = NULL;
}

bool XRingBuffer::pushData(const char* data, const DWORD size)
{
	// 一定要先memcpy再设置writePos,读取也一样,要先有数据,才设置标志
	const DWORD freeSize = getFreeSize();
	if(size > freeSize)
	{
		return false;
	}
	// 此处一定要先置state,如果把置state放在写指针增加完成后,有可能出现在置state之前,数据被读取空了,
	// 写指针和读指针相等,虽然读空时把state置为了Empty,但紧接着此处又错误地置为Full了。
	if(size == freeSize)
	{
		_state = eXRingBufferState_Full;
	}
	if(_write_pos < _read_pos)
	{
		assert(_read_pos - _write_pos >= size);
		memcpy_s(_buffer + _write_pos, _read_pos - _write_pos, data, size);
		_write_pos += size;
	}
	else
	{
		if(_write_pos + size > _size)
		{
			// 尾部溢出
			// 先拷贝一部分长度到尾部,然后在从头开始拷贝
			DWORD wSize = _size - _write_pos;
			memcpy_s(_buffer + _write_pos, wSize, data, wSize);
			// 拷贝剩余部分到头部
			memcpy_s(_buffer, freeSize - wSize, data + wSize, size - wSize);
			_write_pos = size - wSize;
		}
		else
		{
			assert(_size - _write_pos >= size);
			memcpy_s(_buffer + _write_pos, freeSize, data, size);
			_write_pos = (_write_pos + size) % _size;
		}
	}
	return true;
}

bool XRingBuffer::copyData(char* dest, const DWORD destSize, const DWORD size)
{
	if(size > getUsedSize())
	{
		return false;
	}
	if(_read_pos + size > _size)
	{
		// 到尾部还不够,需要再从头读
		// 先把尾部的拷进去
		DWORD wSize = _size - _read_pos;
		memcpy_s(dest, destSize, _buffer + _read_pos, wSize);
		// 拷贝头部的
		memcpy_s(dest + wSize, destSize - wSize, _buffer, size - wSize);
	}
	else 
	{
		memcpy_s(dest, destSize, _buffer + _read_pos, size);
	}
	return true;
}

bool XRingBuffer::popData(char* dest, const DWORD destSize, const DWORD popSize)
{
	const DWORD usedSize = getUsedSize();
	if(popSize > usedSize)
	{
		return false;
	}
	// 此处一定要先置state,原因参考pushData注释
	if(usedSize == popSize)
	{
		_state = eXRingBufferState_Empty;
	}
	if(_read_pos + popSize > _size)
	{
		// 到尾部还不够,需要再从头读
		// 先把尾部的拷进去
		DWORD wSize = _size - _read_pos;
		memcpy_s(dest, destSize, _buffer + _read_pos, wSize);
		// 拷贝头部的
		memcpy_s(dest + wSize, destSize - wSize, _buffer, popSize - wSize);
		_read_pos = popSize - wSize;
	}
	else 
	{
		memcpy_s(dest, destSize, _buffer + _read_pos, popSize);
		_read_pos = (_read_pos + popSize) % _size;
	}
	return true;
}

bool XRingBuffer::popData(const DWORD popSize)
{
	const DWORD usedSize = getUsedSize();
	if(popSize > usedSize)
	{
		return false;
	}
	// 此处一定要先置state,原因参考pushData注释
	if(usedSize == popSize)
	{
		_state = eXRingBufferState_Empty;
	}
	if(_read_pos + popSize > _size)
	{
		// 到尾部还不够,需要再从头读
		DWORD wSize = _size - _read_pos;
		_read_pos = popSize - wSize;
	}
	else 
	{
		_read_pos = (_read_pos + popSize) % _size;
	}
	return true;
}

const DWORD XRingBuffer::getUsedSize() const
{
	return _size - getFreeSize();
}

const DWORD XRingBuffer::getFreeSize() const
{
	if(_write_pos == _read_pos)
	{
		if(_state == eXRingBufferState_Empty)
		{
			return _size;
		}
		else
		{
			return 0;
		}
	}
	else if(_write_pos < _read_pos)
	{
		return _read_pos - _write_pos;
	}
	else 
	{
		return _size - (_write_pos - _read_pos);
	}
}

 

你可能感兴趣的:(IO)