[zeromq-dev] zmq build, examples, packaging

Dirk Eddelbuettel edd at debian.org
Thu Dec 10 16:55:08 CET 2009


On 10 December 2009 at 08:12, Martin Sustrik wrote:
| Dirk Eddelbuettel wrote:
| > Hi Martin,
| > 
| > On 9 December 2009 at 13:34, Martin Sustrik wrote:
| > | Hi Dirk,
| > | 
| > | > Ok. I also quickly packaged Tk-Graph for internal use and am now running with
| > | > freshly built zeromq_1.0.1 packages based on the git repo thanks to Pavol's
| > | > patches.  [ BTW graph.pl didn't get installed along with the exchange/
| > | > examples, it probably should. ]
| > | 
| > | Yes, I would say so.
| > | 
| > | > I currently have two machines '1' and '2'.  They are on the same subnet and
| > | > both four-core cpus, albeit not the very newest,
| > | > 
| > | > I have  '1'   running     zmq_server,  gwt  and   stat | graph.pl
| > | >         '2'   runing      me 
| > | > so that there should be actual cross-machine traffic.
| > | > 
| > | > But that's not what I am seeing.  While the plot for latency is low (sub 500
| > | > usec pers the graph.pl) I have essentially flatlining throughput at 0.  That
| > | > doesn't seem right.  What am I missing?
| > | 
| > | Not sure. What's the output from stat? Does it really report throughput 
| > | of 0 or is it just a problem with graph.pl?
| > 
| > Maybe it is graph.pl.  I let stat run for a bit and tee(1)'ed the result, see
| > below.  Running is again right now. Latency sits steady at 190, then jumped
| > to 2786, came back down again and now jumped to 8476 which is outside the
| > graph. Now down again.
| 
| Tuning the for deterministic latencies is a delicate job. What HW setup 
| are you running the example on?
| 
| > 
| > Throughput always sits at 0-ish values in the chart, I guess it read 20k:
| > 
| > 1=20000 2=19999 3=48668 4=48684 l=174
| > 1=20000 2=19999 3=48668 4=48684 l=174
| > 1=20000 2=19999 3=48668 4=48684 l=174
| > 1=20000 2=19999 3=48668 4=48684 l=174
| > 1=20000 2=19999 3=48668 4=48684 l=174
| > [...]
| 
| As far as I can say you've set message rate to 20000. Try changing the 
| second command line parameter ("orders per second") when running gtw 
| application. Note that when you exceed maximal throughput, latencies 
| will go out of bounds as the messages will delay in queues.

Ack -- that did it. With 10000 messages I am getting more sensible results on
this not-exactly current hardware.

Dirk

-- 
Three out of two people have difficulties with fractions.



More information about the zeromq-dev mailing list