[zeromq-dev] Question related to performance of ZMQ library

[Robert Nogiec]

I want to ask for advice due to my observations.

I am evaluating ZMQ library: I created simple client/server application,
and performed some performance test.

During this test I send million ‘packets’ of size 200, 1000, 5000 and 10000
bytes each (I not use multipart msgs)

I measure time required for send all million packets.

Everything is done on localhost (initially I am interesting in ZMQ as IPC
communication wrapper)

and test was performed on embedded machine MIPS based.

During test I compare solution based on ZMQ (PUSH/PULL, XREQ/XRESP), native
Unix sockets and internal solution based

on UDP.

I unable to understand some results:

I in comparison to UDS socket I have similar results in case small packets
like 200 bytes but significant difference in case

bigger packets (like ~10kB). Additionally what is wondering for me is that
required time not change so much in range

200 – 10000 in case UDS socks and UDP, but in case ZMQ I observed this time
increase almost linear.

I expect worse performance due to double  buffering, encapsulation,
reliability overhead  but I understand why difference is so big in

case bigger data and almost same in case small packets?

Is any possibility to adjust configuration (like buffer size?) to improve
performance in case bigger amount of data?



Regards,



Robert Nogiec
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20151009/37f0e1d1/attachment.htm>