async operations on socket monitors without busy spinning?

[zeromq]

async operations on socket monitors without busy spinning?

[rodgert]

Have you tested with the most recent code? There was an issue fixed
recently that would lead to 100% CPU utilization.

The poll should work the same as the rest of Asio, which will block until
an FD becomes readable, or a periodic timeout occurs.

On Friday, June 19, 2015, David Nadlinger [email protected] wrote:

In my fully asynchronous application, I need to keep track of the number
of connections on some ZeroMQ sockets, for which I am using the socket
monitor feature. However, because the monitoring socket pair uses the
inproc transport, azmq resorts to polling for new events for implementing
async_read. As a side effect, this leads my program to hog 100% of its
CPU core all of the time. Is there any chance to implement this more
efficiently without changing ZeroMQ itself? (I'd like to avoid shipping a
custom fork.)

As a workaround, I currently only check for new socket events on a timer.
But this is obviously not ideal, as it introduces extra latency to the
handling of these events.

—
Reply to this email directly or view it on GitHub
#97.

[dnadlinger]

Sorry for not responding earlier; yes, I believe this has been fixed. I'm a bit hazy on the details, but I checked my code base, and I definitely reverted the timer workaround.