[nanomsg] Re: help with application level message payload
- From: zerotacg <zero@xxxxxxxxxxxxxxxx>
- To: nanomsg@xxxxxxxxxxxxx
- Date: Fri, 21 Feb 2014 07:16:16 +0100
I think you got me wrong, I didn't mean to use nn_allocmsg, quite the
opposite, just create a nn_msghdr yourself, there you can have
multiple "payloads" or buffers to send.
check the description of nn_msghdr.msg_iov and nn_iovec
I'm not quite sure if u can do the same with nn_allocmsg but as I quoted
it's description, nanomsg will only deallocate the message if u set
iov_len to NN_MSG, so just don't, set it as described in the nn_sendmsg
manual to the array size
I hope that helps :D
On 20.02.2014 21:18, Christian Bechette wrote:
> Thanks for the feedback. My ultimate goal is to preserve buffer
> ownership at the application level. I'm a bit puzzled at nn_allocmsg's
> role since even though nn_allocmsg allows me to allocate a buffer, it
> still defers ownership to nanomsg whenever nn_send is called, making it
> impossible to pool its memory. I'm mostly concerned with heap
> fragmentation and such on a media streaming application that runs 24/7.
>
> In an ideal world I would also need to pass multiple
> application-allocated payloads per message, which is not allowed in
> nn_sendmsg. Seems like it's supported at the WSASend level but not
> through the nn api?
>
>
> On Thu, Feb 20, 2014 at 1:54 PM, zerotacg <zero@xxxxxxxxxxxxxxxx
> <mailto:zero@xxxxxxxxxxxxxxxx>> wrote:
>
> what you describe seems to me like default behavior [1] if u allocate
> the message yourself
>
> [quote]
> Alternatively, to send a buffer allocated by nn_allocmsg(3) function set
> iov_base to point to the pointer to the buffer and iov_len to NN_MSG
> constant. In this case a successful call to nn_send will deallocate the
> buffer.
> [/quote]
>
>
> [1] http://nanomsg.org/v0.2/nn_sendmsg.3.html
>
> On 20.02.2014 17:55, Christian Bechette wrote:
> > Hello all, I'm currently using nanomsg for a media streaming
> project. If
> > possible I'd appreciate your input on my current situation;
> >
> > For example in a TCP PAIR scenario, is it possible for nanomsg to
> use my
> > own application level message payload and not allocate a new one every
> > time I do nn_sendmsg? In the case of a blocking send I imagine the io
> > completion port will guarantee that the payload can be discarded after
> > leaving nn_sendmsg.
> >
> > The same applies to the receiving end, is there a way it could write
> > into my own buffers?
> >
> > Thanks for reading :)
> >
> > Christian Bechette
>
>
>
>
>
> --
> Christian.Bechette@xxxxxxxx <mailto:Christian.Bechette@xxxxxxxx>
> Tightrope Media Systems
> (866)-866-4118 x229
Other related posts:
