[rabbitmq-discuss] Poor performance using a single RabbitMQ connection on high-latency networks

Mon Feb 8 13:10:05 GMT 2010

Fascinating...not sure if the subject has been beaten to death yet, but, my
results are somewhat different.

The following are on Windows 7 X64:
------------------------------------------------------------------------------------------------------------------------------------------
Fixed send and receive buffers at 65535:

c:\AMQP\RabbitMQ\rabbitmq-java-client-1.7.1\test\src>java -server
com.rabbitmq.e
xamples.MulticastMain -hlocalhost -p5672 -a -z 60 -i 10 -s 1024 -r 12000
starting consumer #0
starting producer #0
recving rate: 4095 msg/s, min/avg/max latency: 449/489755/1646688
microseconds
sending rate: 4849 msg/s
recving rate: 5395 msg/s, min/avg/max latency: 1566/374119/1485288
microseconds
sending rate: 5226 msg/s
recving rate: 4620 msg/s, min/avg/max latency: 785/299498/1330197
microseconds
sending rate: 5176 msg/s
recving rate: 5137 msg/s, min/avg/max latency: 918/381703/1385686
microseconds
sending rate: 5357 msg/s
recving rate: 5873 msg/s, min/avg/max latency: 379/445433/1677491
microseconds
sending rate: 5192 msg/s
recving rate: 5274 msg/s, min/avg/max latency: 426/244059/1061180
microseconds
recving rate avg: 5193 msg/s
sending rate avg: 5211 msg/s
------------------------------------------------------------------------------------------------------------------------------------------
This is with no buffers specified on the same machine:

c:\AMQP\RabbitMQ\rabbitmq-java-client-1.7.1\test\src>java -server
com.rabbitmq.e
xamples.MulticastMain -hlocalhost -p5672 -a -z 60 -i 10 -s 1024 -r 12000
starting consumer #0
starting producer #0
recving rate: 3302 msg/s, min/avg/max latency: 6666/493540/1274826
microseconds
sending rate: 3461 msg/s
recving rate: 3696 msg/s, min/avg/max latency: 6462/270603/1022742
microseconds
sending rate: 3688 msg/s
recving rate: 3684 msg/s, min/avg/max latency: 707/903615/2256495
microseconds
sending rate: 3836 msg/s
recving rate: 3923 msg/s, min/avg/max latency: 801/387877/1060920
microseconds
sending rate: 4052 msg/s
recving rate: 3594 msg/s, min/avg/max latency: 2566/253710/1099121
microseconds
sending rate: 3473 msg/s
recving rate: 3900 msg/s, min/avg/max latency: 2045/308265/658904
microseconds
recving rate avg: 3708 msg/s
sending rate avg: 3715 msg/s

------------------------------------------------------------------------------------------------------------------------------------------
On Mon, Feb 8, 2010 at 13:24, Matthew Sackman <matthew at lshift.net> wrote:

> On Sat, Feb 06, 2010 at 04:32:35AM +0100, Holger Hoffstaette wrote:
> > This rabbit hole goes deep..
>
> Indeed it does. I repeated all my tests this morning with the buffers
> all hard set to 65536 and things got slightly better with large
> messages (~1kB), but worse with small messages.
>
> I then set up a high latency loopback:
> # tc qdisc add dev lo root netem delay 100ms 20ms distribution pareto loss
> 0.01%
>
> So a 100ms delay, ±20ms on a pareto distribution, with 0.01% packet
> loss.
>
> With our default buffer sizes:
> rabbitmq-java-client/build/dist$ sh runjava.sh
> com/rabbitmq/examples/MulticastMain -a -z 60 -i 10 -s 1024 -r 12000
> starting consumer #0
> starting producer #0
> recving rate: 8 msg/s, min/avg/max latency: 207432/3279948/6437079
> microseconds
> sending rate: 322 msg/s
> recving rate: 13 msg/s, min/avg/max latency: 6437120/11372211/16522991
> microseconds
> sending rate: 14 msg/s
> recving rate: 14 msg/s, min/avg/max latency: 16522969/21456014/26551356
> microseconds
> sending rate: 13 msg/s
> recving rate: 13 msg/s, min/avg/max latency: 26551374/31650558/36685934
> microseconds
> sending rate: 13 msg/s
> recving rate: 13 msg/s, min/avg/max latency: 36687449/41818269/46842999
> microseconds
> sending rate: 13 msg/s
> recving rate: 13 msg/s, min/avg/max latency: 46851173/51931717/56934090
> microseconds
> recving rate avg: 13 msg/s
> sending rate avg: 67 msg/s
>
> With buffers set to 65536 and 65536:
> recving rate: 180 msg/s, min/avg/max latency: 220489/3300700/6318430
> microseconds
> sending rate: 648 msg/s
> recving rate: 286 msg/s, min/avg/max latency: 6318355/10433052/12858930
> microseconds
> sending rate: 256 msg/s
> recving rate: 264 msg/s, min/avg/max latency: 12676703/13635080/14137168
> microseconds
> sending rate: 294 msg/s
> recving rate: 282 msg/s, min/avg/max latency: 12437468/13308885/14238964
> microseconds
> sending rate: 275 msg/s
> recving rate: 272 msg/s, min/avg/max latency: 12971076/13367869/13966905
> microseconds
> sending rate: 262 msg/s
> recving rate: 251 msg/s, min/avg/max latency: 13644724/14051244/14746586
> microseconds
> recving rate avg: 265 msg/s
> sending rate avg: 332 msg/s
>
> With no buffers set (thus allowing auto resizing to take place):
> recving rate: 1012 msg/s, min/avg/max latency: 196813/2498703/3712777
> microseconds
> sending rate: 6146 msg/s
> recving rate: 6104 msg/s, min/avg/max latency: 2782802/4983136/5602666
> microseconds
> sending rate: 10466 msg/s
> recving rate: 9649 msg/s, min/avg/max latency: 5440633/5971475/6431049
> microseconds
> sending rate: 10637 msg/s
> recving rate: 9970 msg/s, min/avg/max latency: 6204889/6718248/7106264
> microseconds
> sending rate: 6681 msg/s
> recving rate: 9535 msg/s, min/avg/max latency: 3934919/4661861/7229773
> microseconds
> sending rate: 10795 msg/s
> recving rate: 8960 msg/s, min/avg/max latency: 4055794/4703495/5892391
> microseconds
> recving rate avg: 7755 msg/s
> sending rate avg: 9121 msg/s
>
> Needless to say, my flabber has been ghasted. Many thanks for bringing
> this issue to our attention. Unless there are further revelations, it's
> very likely we'll remove this setting in the next release.
>
> Matthew
>
> _______________________________________________
> rabbitmq-discuss mailing list
> rabbitmq-discuss at lists.rabbitmq.com
> http://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss
>

-- 
---
John Apps
(49) 171 869 1813
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20100208/1e8e8d6d/attachment.htm 