Martin,

most of my problems got solved by

a) using nanosleep() - stopped the cpu spikes
b) in the subscriber receive loop switching to zmq_poll() and doing a
msg_init and msg_close before and after the zmq_recv()
c) on the publisher side, using msg_init_data() with a noop_free() that
allows me to reuse the data buffer again and overwrite the fields for msg
index and send timestamp

however there is still a subtle monotonic memory growth that I'm seeing on
the subscriber side. the messages on the sub have been drained and even
msg_close'd.

in an hour's run, the subscriber process (with 4 sub sockets and receiving a
1000 msgs/sec of msg size: 2048 bytes) memory monotonically grows to about
1Mb from starting memory footprint of 200 Kb. even after i kill the
publisher process, the subscriber process memory doesn't come down.

I strongly suspect this might be a bug in my benchmark code and will share
the minimal testcase, once I've valgrind'd the heck out of that piece of
code :)

--Saurav
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

Reply via email to