qpid-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alan Conway <acon...@redhat.com>
Subject Re: How to test the performance quid c++ broker
Date Thu, 24 Jul 2014 12:59:36 GMT
On Thu, 2014-07-24 at 13:24 +0800, 郑勰 wrote:
> Hi, Alan
> 	Thanks, I execute the same order with yours .
> 	
> 	qpid-cpp-benchmark --repeat 3 --summariz -q 6 -s 3 -r 3 -m 10000
> 	
> send-tp	recv-tp	l-min	l-max	l-avg	total-tp
> 34197	28770	38.60	1452.75	881.22	26933
> 34926	30274	42.38	1370.06	743.35	28669
> 35381	30359	34.67	1277.13	683.16	28410
> 
> 	It is far lower than your result.   I wonder what your PC’s configuration is . My
server is with 8 cpus.

mrg32.lab.bos.redhat.com 2.6.32-358.el6.x86_64 x86_64: 15950Mb 2493MHz
8-core/2-cpu
  Intel(R) Xeon(R) CPU           E5420  @ 2.50GHz
  Red Hat Enterprise Linux Server release 6.4 (Santiago)

Very important point I forgot to mention: are you doing a release
build? 

cmake -DCMAKE_BUILD_TYPE=Release

That makes a big difference. It enables optimization flags for the C++
compiler. The default is not optimized.


> 
> 
> 在 2014年7月24日,上午1:56,Alan Conway <aconway@redhat.com> 写道:
> 
> > On Tue, 2014-07-22 at 10:04 +0800, 郑勰 wrote:
> >>> Hi, Alan
> >> 
> >> Thanks, I know little about python, when I execute 
> >> 
> >> 	./qpid-cpp-benchmark -q 1 -s 1 -r 1 -m 200000 --summarize --repeat 5
> >> 
> >> This is the result:
> >> 	
> >> send-tp		recv-tp		l-min	l-max		l-avg		total-tp
> >> 8150	8127	0.90	85.12	33.34	8112
> >> 8149	8134	0.75	91.78	33.08	8119
> >> 8118		8101	0.73	94.25	32.94	8087
> >> 8104	8070	0.85	105.40	34.06	8055
> >> 8079	8046	0.82	102.83	33.72	8031
> >> 
> >> send-tp is the toppest messages per second,  what does it mean of l-max and
total-to?    
> > 
> > send-tp is the number of messages per second sent by the sender, added
> > if there are many senders and you use --summarize.
> > 
> > recv-tp is the number of messages per second received by the receiver,
> > also added if many receivers.
> > 
> > total-tp is the total end-to-end throughput in messages per second, if
> > you have more than one sender/receiver/queue this is usually a little
> > lower than the individual tps. It calculates the total throughput over
> > the period of the first message sent by any sender to the last message
> > received by any receiver. 
> > 
> > l-min/max/avg are the minimum, maximum and average latency in
> > milliseconds.
> > 
> >> 
> >> From the result, My boss said to me:”Hnnn,  not good …”, well, a blue
day …
> >> 
> >> So I want to know your result, I expect send-to and recv-tp should be at least
200,000.
> > 
> > That will depend a lot on hardware. Also bear in mind that you can get
> > much higher total throughput when you have multiple senders and
> > receivers since the broker can process messages in parallel.
> > 
> > For example:
> > 
> > aconway@mrg32 release (trunk)$ qpid-cpp-benchmark --repeat 3 --summarize
> > -q 1 -s 1 -r 1 -m 10000
> > send-tp recv-tp l-min   l-max   l-avg   total-tp
> > 26847   26630   0.43    13.49   5.80    25150
> > 25998   25437   0.35    18.67   9.46    24056
> > 27763   27728   0.15    15.01   5.44    26185
> > 
> > aconway@mrg32 release (trunk)$ qpid-cpp-benchmark --repeat 3 --summarize
> > -q 6 -s 3 -r 3 -m 10000
> > send-tp recv-tp l-min   l-max   l-avg   total-tp
> > 96582   92764   6.50    229.86  93.26   80262
> > 97851   92901   6.12    229.14  91.35   81030
> > 99183   95358   5.89    224.20  90.07   81362
> > 
> > There are a number of parameters you can adjust, check qpid-send --help
> > and qpid-receive --help for details. You can run qpid-cpp-benchmark with
> > --send-arg and --receive-arg to pass arguments to the senders and
> > receivers.
> > 
> >> 
> >> 
> >> 
> >> 在 2014年7月22日,上午12:29,Alan Conway <aconway@redhat.com>
写道:
> >> 
> >>> On Mon, 2014-07-21 at 11:45 +0200, Jakub Scholz wrote:
> >>>> Hi,
> >>>> 
> >>>> Qpid project contains two utilities: qpid-perftest and qpid-latency-test.
> >>>> You can use these instead of your own program. If you use these you
can
> >>>> share the complete command which you used to start the performance test.
> >>>> Right now it is not clear for example how big your messages were or
what
> >>>> was the configuration of your receiver. So it is hard to judge the
> >>>> performance.
> >>>> 
> >>>> Regards
> >>>> Jakub
> >>> 
> >>> Those utilities are useful but a bit out of date - they use deprecated
> >>> APIs. 
> >>> 
> >>> You should take a look at qpid-send, qpid-receive and
> >>> qpid-cpp-benchmark. qpid-send and receive are quite flexible
> >>> general-purpse test tools for sending and receiving messages.
> >>> 
> >>> qpid-cpp-benchmark is a python script that runs multiple instances of
> >>> qpid-send and qpid-receive in a variety of configurations, and collects
> >>> latency and throughput results. 
> >>> 
> >>> Cheers,
> >>> Alan
> >>> 
> >>> 
> >>> ---------------------------------------------------------------------
> >>> To unsubscribe, e-mail: users-unsubscribe@qpid.apache.org
> >>> For additional commands, e-mail: users-help@qpid.apache.org
> >>> 
> >> 
> > 
> > 
> > 
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: users-unsubscribe@qpid.apache.org
> > For additional commands, e-mail: users-help@qpid.apache.org
> 



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@qpid.apache.org
For additional commands, e-mail: users-help@qpid.apache.org


Mime
View raw message