Platform-Specific Implementation Notes

This section lists platform-specific implementation details, such as the default demultiplexing mechanism, the number of threads created internally, and when threads are created.

Linux Kernel 2.4

Demultiplexing mechanism:

Uses select for demultiplexing. This means that the number of file descriptors in the process cannot be permitted to exceed FD_SETSIZE.

Threads:

Demultiplexing using select is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

Linux Kernel 2.6

Demultiplexing mechanism:

Uses epoll for demultiplexing.

Threads:

Demultiplexing using epoll is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

Linux Kernel 5.10

Demultiplexing mechanism:

If BOOST_ASIO_HAS_IO_URING is defined, uses io_uring for file-related asynchonous operations.
Uses epoll for demultiplexing other event sources.
Optionally uses io_uring for all asynchronous operations if, in addition to BOOST_ASIO_HAS_IO_URING, BOOST_ASIO_DISABLE_EPOLL is defined to disable epoll.

Threads:

Demultiplexing using epoll and io_uring is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

Solaris

Demultiplexing mechanism:

Uses /dev/poll for demultiplexing.

Threads:

Demultiplexing using /dev/poll is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

QNX Neutrino

Demultiplexing mechanism:

Uses select for demultiplexing. This means that the number of file descriptors in the process cannot be permitted to exceed FD_SETSIZE.

Threads:

Demultiplexing using select is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

Mac OS X

Demultiplexing mechanism:

Uses kqueue for demultiplexing.

Threads:

Demultiplexing using kqueue is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

FreeBSD

Demultiplexing mechanism:

Uses kqueue for demultiplexing.

Threads:

Demultiplexing using kqueue is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

AIX

Demultiplexing mechanism:

Uses select for demultiplexing. This means that the number of file descriptors in the process cannot be permitted to exceed FD_SETSIZE.

Threads:

Demultiplexing using select is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

HP-UX

Demultiplexing mechanism:

Uses select for demultiplexing. This means that the number of file descriptors in the process cannot be permitted to exceed FD_SETSIZE.

Threads:

Demultiplexing using select is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

Tru64

Demultiplexing mechanism:

Uses select for demultiplexing. This means that the number of file descriptors in the process cannot be permitted to exceed FD_SETSIZE.

Threads:

Demultiplexing using select is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

At most min(64,IOV_MAX) buffers may be transferred in a single operation.

Windows 95, 98 and Me

Demultiplexing mechanism:

Uses select for demultiplexing.

Threads:

Demultiplexing using select is performed in one of the threads that calls io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

For sockets, at most 16 buffers may be transferred in a single operation.

Windows NT, 2000, XP, 2003, Vista, 7, 8, 10 and 11

Demultiplexing mechanism:

Uses overlapped I/O and I/O completion ports for all asynchronous socket operations except for asynchronous connect.
Uses select for emulating asynchronous connect.

Threads:

Demultiplexing using I/O completion ports is performed in all threads that call io_context::run(), io_context::run_one(), io_context::poll() or io_context::poll_one().
An additional thread per io_context is used to trigger timers. This thread is created on construction of the first basic_deadline_timer or basic_waitable_timer objects.
An additional thread per io_context may be used for select demultiplexing. This thread is created on the first call to:
- A socket async_wait() function, except when using wait_read on a stream-oriented socket. (For wait_read on a stream-oriented socket, the overlapped I/O operation WSARecv is used and no additional thread is required.)
- A socket async_connect() operation, if the overlapped I/O operation ConnectEx is unavailable. (On recent versions of Windows, ConnectEx is used and no additional thread is required.)
An additional thread per io_context is used to emulate asynchronous host resolution. This thread is created on the first call to either ip::tcp::resolver::async_resolve() or ip::udp::resolver::async_resolve().

Scatter-Gather:

For sockets, at most 64 buffers may be transferred in a single operation.
For stream-oriented handles, only one buffer may be transferred in a single operation.

Boost.Asio provides limited support for the Windows Runtime. It requires that the language extensions be enabled. Due to the restricted facilities exposed by the Windows Runtime API, the support comes with the following caveats:

The core facilities such as the io_context, strand, buffers, composed operations, timers, etc., should all work as normal.
For sockets, only client-side TCP is supported.
Explicit binding of a client-side TCP socket is not supported.
The cancel() function is not supported for sockets. Asynchronous operations may only be cancelled by closing the socket.
Operations that use null_buffers are not supported.
Only tcp::no_delay and socket_base::keep_alive options are supported.
Resolvers do not support service names, only numbers. I.e. you must use "80" rather than "http".
Most resolver query flags have no effect.

Demultiplexing mechanism:

Uses the Windows::Networking::Sockets::StreamSocket class to implement asynchronous TCP socket operations.

Threads:

Event completions are delivered to the Windows thread pool and posted to the io_context for the handler to be executed.
An additional thread per io_context is used to trigger timers. This thread is created on construction of the first timer objects.

Scatter-Gather:

For sockets, at most one buffer may be transferred in a single operation.