Design Internals of UNIX Non Blocking Sockets

Fundamental of asynchronous programming: Non Blocking Sockets

Why async socket programming

• An application can create thousands of sockets but creating a thread per socket becomes infeasible for a large number (e.g. 100k sockets)
• An async framework consists non-blocking sockets and an event loop

  * The kernel allows applications to seek for an event
  * Application registers a callback to be called when the event occur
  * Callback runs by the same application thread after kernel intimates
    that an event happened on a socket
  

• This model uses a single thread to poll many sockets for various events and calls the callback.
• It is concurrent that is runs many tasks but not parallel execution of tasks.

How do a socket turn into non-blocking?

• A socket is a file, read and written with standard POSIX calls.
• A system call for read operation uses the kernel BIO (Block I/O). A BIO provides an interface to access a block device.
  • ioctl allows to set a fd non-blocking for read/write calls
  • Learn all about BIO: https://lwn.net/Articles/736534/

  FIONBIO int
  Set non-blocking I/O mode if the argument is non-zero. In non-
  blocking mode, read(2) or write(2) calls return -1 and set errno
  to EAGAIN immediately when no data is available (This is
  equivalent to fcntl() F_SETFL O_NONBLOCK and the fcntl() form
  should be preferred)

Written with StackEdit.