[zeromq-dev] Why ZeroMQ has so much overhead?

Julie Anderson julie.anderson.uk at gmail.com
Wed Aug 29 05:04:59 CEST 2012


I sent an email earlier titled: "Java NIO Selector Minimum Possible
Latency".

I am NOT doing anything fancy like kernel bypass.


On Tue, Aug 28, 2012 at 8:17 PM, Steven McCoy <steven.mccoy at miru.hk> wrote:

> On 28 August 2012 20:11, Julie Anderson <julie.anderson.uk at gmail.com>wrote:
>
>> Over 10GbE ZeroMQ claims a latency of 33 micros from end-to-end<http://www.zeromq.org/results:10gbe-tests-v031>.
>> If you subtract 2 micros from "over-the-wire" transit time it is still 31
>> micros. That's a lot!
>>
>> But if you make a simple test sending one packet from one side to the
>> other, the total time is around 11 micros, or three times faster.
>>
>> What does ZeroMQ do to introduce 3 times more latency to the messages?
>>
>>
> That document is 4 years old, hardware is faster today and low-microsecond
> RTT requires a user-space stack to bypass kernel overheads which Mellanox
> didn't have then.
>
>
> http://www.mellanox.com/content/pages.php?pg=products_dyn&product_family=106&menu_section=69
>
> --
> Steve-o
>
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20120828/ae146241/attachment.htm>


More information about the zeromq-dev mailing list