[zeromq-dev] On the performance of ZMQ_STREAM sockets

André Caron andre.l.caron at gmail.com
Wed Jan 22 04:33:15 CET 2014


BTW, I'm going to experiment with a ZeroMQ-based backend infrastructure
that supports credit-based flow control (I'm not only interested in the
stream sockets).  I'll look at that once I've had all my fun with the HTTP
stuff :-)

André


On Tue, Jan 21, 2014 at 10:30 PM, André Caron <andre.l.caron at gmail.com>wrote:

> FYI,
>
> I just wrote a small HTTP server based on ZMQ_STREAM sockets and an
> outdated version of http-parser[1] wrapped in some C++ objects with naïve
> buffering.
>
> I profiled it with "ab" and got some interesting results:
> - ~6K req/sec without HTTP keep alive; and
> - ~19.5K req/sec with HTTP keep alive.
>
> This is a super naïve server, there's lots more to add in there that will
> likely affect this performance, but it goes a long way towards
> demonstrating the incredible job you guys have done with this library :-D
>
> I have a few projects up on GitHub that implement some HTTP-related
> protocols (WebSockets, FastCGI, SCGI, etc.) in a way that are independent
> from the IO engine (FSMs for the win).  I've been experimenting with
> several IO engines for C/C++ and hadn't yet found something that was simple
> enough to let me focus on the interesting bits.  These tests prove that the
> search is finally over :-)
>
> I'll be wrapping that up in a ZeroMQ-based HTTP gateway in the upcoming
> weeks.  I'll post a link to the GitHub project as soon as I post it up
> (should be very soon).
>
> [1]: https://github.com/joyent/http-parser
>
> Cheers,
>
> André
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20140121/8a979b62/attachment.htm>


More information about the zeromq-dev mailing list