[zeromq-dev] Pub/Sub scalability over TCP vs epgm/pgm

Colin Ingarfield colin at ingarfield.com
Mon Apr 9 23:15:35 CEST 2018



On 4/9/18 4:09 PM, Mark Delany via zeromq-dev wrote:
>>> can have topics around 500 and subscribers could be from 500 to 1000.
>> The only way to know is to test it
> This is true and also measure your system while testing to identify
> potential system bottlenecks. What else is your application doing
> besides making zeromq calls? Is there I/O involved?
>
> You say you are getting delays. How large are these? How busy are the
> systems when you're getting these delays.
>
> As a general rule, 1,000 or so TCP sockets is no big deal for a modern
> Unix/Linux kernel. I would also be surprised if 500 topics causes a
> much delay. Even a linear search per message would happen pretty
> quickly on a modern CPU. I am assuming that zeromq does better than
> linear since topics are exact prefix match there are plenty of
> alogrithmic solutions that are far faster than O(n).
I believe it uses a prefix matching trie.  Worst case is O(length of
longest topic).  If topic names are reasonably short, it's negligible on
a modern cpu.

> A number of vmstat samples will give a general idea of what's going
> on.
>
>
> Mark.
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> https://lists.zeromq.org/mailman/listinfo/zeromq-dev




More information about the zeromq-dev mailing list