[zeromq-dev] 0MQ multicast message lost?

[ntupel at googlemail.com]

On Thu, Oct 7, 2010 at 5:16 PM, Steven McCoy <steven.mccoy at miru.hk> wrote:
> On 7 October 2010 23:06, <ntupel at googlemail.com> wrote:
>>
>> On Thu, Oct 7, 2010 at 4:20 PM, Steven McCoy <steven.mccoy at miru.hk> wrote:
>> > You are setting the rate limit at 64kbs
>>
>> I did this:
>>
>> int64_t rate = 64000;
>> receiver.setsockopt(ZMQ_RATE, &rate, sizeof(rate));
>>
>> According to man zmq_setsockopt the rate unit is kilobits per second
>> with the default being 100. So my impression was I was setting it to
>> 62.5 mbps. Using 65536000 for example didn't make a difference.
>>
>
> Oh ok, 0MQ multiplies up, I have measured (8 * 10 * 1000) to be pretty
> accurate to 80Mb/s.
> What kind of network are you running this on?

It is a 100Mbit network. I am currently experimenting with 0MQs
performance test programs
(http://github.com/zeromq/zeromq2/blob/master/perf/local_thr.cpp, and
http://github.com/zeromq/zeromq2/blob/master/perf/remote_thr.cpp). I
just set ZMQ_MCAST_LOOP to 0 and try different rate values. It seems
that with 10000 messages of size 255 bytes it is mostly fine for rate
<= 1024 (but still not always). Bigger values cause it to not receive
all messages. To measure bulk throughput I did a scp of a 50MB file
which  was using 10MB/s. So I am puzzled why the rate value has to be
that low. After all this is just about 488 messages per second. Is it
really possible that there is now room for RDATA? What else might be
the problem?

Many thanks for all your help!

-nt