PACKET_MMAP PF_RING 的主要区别在哪

http://blog.csdn.net/xulai1001/article/details/7356582 

http://www.mjmwired.net/kernel/Documentation/networking/packet_mmap.txt

看PACKET_MMAP的介绍，mmap 一个ring buffer到用户空间 poll 读取，减少系统调用和内存拷贝。

PF_RING也是使用mmap ring buf, poll  下面mark web上面二者比较的一些讨论？

  [Ntop-misc] PF_RING or PACKET_MMAP?  (一个讨论两者的zero copy机制差别的帖子)

    

  Comparison between packet_mmap and pf_ring  (一个讨论比较两者基础性能帖子,说pf_ring没有DNA和修改驱动仍然更快)

   

  From the kernel to the user space (DMA)

   

I read your blog post, and it seems that PACKET_MMAP, is almost the same idea, that the one I proposed (mapping the DMA directly in the user-land), however, mapping the DMA directly in the userland makes it a real 0 copy (that's what I understand) while using PACKET_MMAP is only 0 copy from the kernel to the userland, but because you still have a copy from the DMA to the ring buffer of PACKET_MMAP, no ? but the advantage of PACKET_MMAP, is that you do not need to rewrite any driver. – 

  Improving Passive Packet Capture: Beyond Device Polling

    

    http://blog.csdn.net/aaa6695798/article/details/4456672

目前能看到的是， PACKET_MMAP只支持内核和用户态之间zero copy，但是内核里面还有一次ring buffer到DMA拷贝；

而PF_RING 通过DNA支持真正的zero copy，具体实现方案有待进一步研究，RTFS

---------------------------------------------------------------------------------

相关研究讨论的帖子

源自BSD的类似技术netmap