Re: AIO sockets

On 29 Feb., 21:33, Jo <JoJoTwill...@xxxxxxxxxxx> wrote:
Is there a tutorial somewhere on how to use AIO for sockets?http://www.kegel.com/c10k.htmlseems to be very complete with no
information, suggesting that it can be done, but trying their links
just adds to my frustration. One thing I find very confusing is that
it seems to talk about polling in parallel with asynchronous I/O.
Doesn't polling defeat the purpose of asynchronism?

Well i guess that most implementations on Linux still sucks. Thats why
aio_ is still excluded from LSB even if LSB defines the rt library.
Last time i looked (okay long ago) aio was just implemented on the
user level abstraction with starting threads. So it makes sense to
believe that AIO on sockets is implemented with a background thread
that is polling on different sockets (and a control file descriptor).
I've implemented something equal in my Eiffel runtime. And yes, it
removes a lot of the optimization that could be done with kernel
supported AIO.
.

Relevant Pages

  • Re: aio_read/write versus O_NONBLOCK
    ... pipe, or socket--are line based, or otherwise based on content, not buffer ... nearly as much to sockets. ... Another thread can invalidate readiness by exhausting a buffer. ... The OS can complete the AIO ...
    (comp.unix.programmer)
  • Asynchronous IO (aio_read, aio_write, etc) and sockets
    ... sockets instead of select/poll, read and write? ... I/O but there is no explicit declaration that it cannot be used for ... One more question is about aio internals. ... Does it mean that actually IO request is performed synchronously by ...
    (comp.unix.aix)
  • Re: AIO sockets
    ... it seems to talk about polling in parallel with asynchronous I/O. ... aio_ is still excluded from LSB even if LSB defines the rt library. ... removes a lot of the optimization that could be done with kernel ... asynchronous I/O that appears to be available today ...
    (comp.os.linux.development.apps)