Well I could but I don't understand why individually sending messages, and
waiting for each before sending the next, would make each individual message go
out slower than if I send a million at once. Wait a min, is it possible I need
to flush the send buffer when I send one message at a time?
--
Rohan Bedarkar
MS, Computer Science
University of Chicago
[email protected]
________________________________
From: Apostolis Xekoukoulotakis <[email protected]>
To: Rohan Bedarkar <[email protected]>; ZeroMQ development list
<[email protected]>
Sent: Monday, October 22, 2012 4:14 PM
Subject: Re: [zeromq-dev] TCP zeromq performance review
Increase the number of msgs(1000000+) and retest. I got 12us.
2012/10/22 Rohan Bedarkar <[email protected]>
Hi Chuck and Pieter
>
>
>I ran the ZMQ perf tests and also made changes to my own tests. Here's the
>outputs. My numbers match the perf tests.
>
>
>Doesn't it seem a little too slow, 40 us for one way tcp message?
>
>ZMQ Latency Test:
>---------------------------
>
>
>./remote_lat tcp://10.34.253.10:20000 10 10000
>message size: 10 [B]
>roundtrip count: 10000
>average latency: 43.045 [us]
>
>
>
>
>ZMQ Throughput Tests
>----------------------------------
>
>
>./local_thr tcp://eth0:20000 10 1000000
>
>
>Total time: 308 [ms]
>message size: 10 [B]
>message count: 1000000
>mean throughput: 3244646 [msg/s]
>mean throughput: 259.572 [Mb/s]
>
>
>My Throughput Test:
>------------------------------
>
>
> Transport: ZMQ
>
> Messages recd: 1000000
> Total Time: 721174 us
> Throughput: 166.395 Mbps
> Avg per msg: 0.721174 us
>
>
>
>My Latency Test:
>-------------------------
>
>
> Transport: ZMQ
>
> Messages sent: 10000
> Average / message: 40.2781 us
> Max time: 65 us
> Min time: 14 us
>
>
>
>
>--
>
>Rohan Bedarkar
>MS, Computer Science
>
>University of Chicago
>[email protected]
>
>
>
>
>
>
>
>
>
>
>________________________________
> From: Chuck Remes <[email protected]>
>To: Rohan Bedarkar <[email protected]>; ZeroMQ development list
><[email protected]>
>Sent: Friday, October 19, 2012 4:49 PM
>Subject: Re: [zeromq-dev] TCP zeromq performance review
>
>
>
>Try it again with the example program local_thr and remote_thr and see if the
>results match using the same inputs.
>
>
>BTW, it looks reasonable to me. I have a similar machine (it's about 4 years
>old) and get around 55 usec latency as an average. Throughput is usually a bit
>higher especially for loopback.
>
>
>cr
>
>
>On Oct 19, 2012, at 4:28 PM, Rohan Bedarkar wrote:
>
>Hi Guys,
>>
>>
>>I've built a performance monitoring framework in C++ over OpenSuse Linux. Its
>>a pretty solid machine with 16 cores.
>>The application publishes TCP messages in one thread and subscribes to them
>>in another. Each message is roughly 10 bytes. Here's the output.
>>
>>
>>Just want to run by these numbers.. What do you think?
>>
>>
>>
>>
>>Bound TCP server to port tcp://*:20000
>>Sending 1000000 msgs
>>
>>Received all messages
>>
>>Analyzing 1000000 msgs..
>>
>>---------------------------------------------------------
>>
>> Performance Report
>>---------------------------------------------------------
>>
>> Transport: ZMQ
>> Messages sent: 1e+06
>> Total Time: 1004.05 ms
>> Average / message: 65.1254 us
>> Max time: 5347 us
>> Min time: 8.00014 us
>> Throughput: 119.515 Mbps
>>
>>---------------------------------------------------------
>
>
>_______________________________________________
>zeromq-dev mailing list
>[email protected]
>http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
>
>
>_______________________________________________
>zeromq-dev mailing list
>[email protected]
>http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
>
--
Sincerely yours,
Apostolis Xekoukoulotakis
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev
