PUB/SUB+PUSH/PULL Messaging Broker - How to improve latency and performance?

Question

I have implemented this ZMQ Client-Server architecture with the order to one message from Sender client, this message arrives at other clients Receiver 1, Receiver 2 and Receiver 3.

This means that the all receivers can see the same data from Sender

In this architecture, I have python clients and c++ clients. The server is coded with Python using pyzmq binding.

The question is: Why on my python clients the message appears faster than on my c++ clients?

Is this related to the fact that the server was coded in Python?

What kind of non-functional considerations should I have, so as to improve the performance of the messaging system?

Is possible think in one thread by each message routed to the receiver?

I would meet your considerations about it

Mathieu

" Why...message appears faster than on c++ clients?" would you mind to put quantitative data about the faster part, what have you measured in [us] + details, about such an experiment setup? The total number of .connect()-ed peers, their respective transport-classes used for that, hardware-colocation v/s distributed-system global layout ( hosts+connections+distributed workloads on each peer-hosting device ), MCVE for each...all that matters, once you ask others to deductively answer why something happens at your desk. That is all needed and fair, isn't it? – Dixon

@Dixon You're right! It's true, Is necessary all tests that you tell me. My question was oriented in relation to possibly the fact that the server is made on python and clients on c++ did impact in the performance, but this would be more transparent and the correct is make some tests about of networking conditions, and workloads on each peer-hosting service. Any consideration about of how to start with these experiments. How to can I measure the arrival time of messages on each peer when these messages departure from their source or origin peer. I appreciate your comments and ideas – Mathieu