If you send and recv 10K messages in a batch, you'll pay a lot for
start-up costs and very little for the actual traffic.

You should send enough messages to keep the test running for 2-10
seconds, then measure the throughput in messages per second.

Then, test on small messages (10 bytes) and large ones (1K or 10K) to
see the effect of memory copying.

-Pieter

On Tue, Apr 16, 2013 at 11:10 AM, crocket <[email protected]> wrote:
> JeroMQ is both a java reimplementation and a higher level API of libzmq.
> labview-zmq is basically a wrapper around libzmq for LabVIEW.
> pyzmq is also basically a wrapper around it.
>
> I made ROUTER.py and DEALER.py and executed them to measure performance.
> Those python codes are attached to this email.
>
> With pyzmq, it takes 200~280ms for DEALER to send 10000 requests to ROUTER
> and retrieve 10,000 replies in batch.
>
> With labview-zmq, the same code takes about 1500~1700ms.
>
> With JeroMQ, it takes about 30~40ms.
>
> Why do different bindings have different performance characteristics?
>
> _______________________________________________
> zeromq-dev mailing list
> [email protected]
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

Reply via email to