RE: [javagroups-users] Performance tests

[Peter E. <pe...@ne...>]

I figured out the problem.

I did the following changes and it improved greatly:
-dedicated a switch to the test, so the only nodes on the switch are the
ones involved in the test.
-made static name mappings to the other servers in the hosts file (we were
using WINS before)

I will perform a more detailed analysis with this setup and report my
findings.

Thanks
-peter 


<signature>
  <name>Peter Eacmen</name>
  <email>pe...@ne...</email>
  <title>Software Engineer</title>
  <affiliation>www.nexaweb.com</affiliation>
  <contact>
    <email>pe...@ne...</email>
    <desk>(617)577-8100 x2019</desk>
    <cell>(617)548-5384</cell>
  </contact>
</signature>

-----Original Message-----
From: jav...@li...
[mailto:jav...@li...] On Behalf Of Peter
Eacmen
Sent: Thursday, January 22, 2004 3:48 PM
To: jav...@li...
Subject: RE: [javagroups-users] Performance tests

No, the config files are consistent for sender and receivers.

Attached are my send.txt, recv.txt, and fc.xml.

TIA for any insight into this issue.


<signature>
  <name>Peter Eacmen</name>
  <email>pe...@ne...</email>
  <title>Software Engineer</title>
  <affiliation>www.nexaweb.com</affiliation>
  <contact>
    <email>pe...@ne...</email>
    <desk>(617)577-8100 x2019</desk>
    <cell>(617)548-5384</cell>
  </contact>
</signature>

-----Original Message-----
From: Bela Ban [mailto:be...@ya...]
Sent: Thursday, January 22, 2004 3:29 PM
To: Peter Eacmen
Cc: jav...@li...
Subject: Re: [javagroups-users] Performance tests

Hi Peter,

did you set the # of senders plus the # of members to be the same in all
{receiver/sender}.txt ? Otherwise, you might expect 8000 msgs, but only get
4000, and would time out, distorting the stop time, and thus all numbers.

I'm working on an improvement to that perf test, which will run over (a)
JGroups, (b) UDP, (c) TCP and (d) JMS.

I recently got 6000 1K msgs/sec on a 4 node cluster, 100Mbps switched
ethernet, 2-cpu compaq proliant boxes. Your number is definitely off...

Peter Eacmen wrote:

> I am trying to replicate the performance tests that were reported on 
> the jgroups website. However I am getting results that are much worse 
> than reported.
>
> Here is my setup:
> 4 homogeneous Win2k Advanced Servers
> -1GB RAM
> -P3 650Mhz Dual Processor
> 100Mbit Switched network
> JDK 1.4.1_02
> Fc.xml file that is shipped in jar file
>
> Here are the results I am getting for the 10000b test:
> expected_msgs=8000, elapsed_time=200722 Received 8000 msgs. in 200722 
> msec.
> Throughput: 39.85611940893375 [msgs/sec]
> Throughput: 398.56119408933745 [KB/sec] Total received: 80.0 [MB]
>
> Which is about 1/10 of the performance reported on the website.
>
> The only difference is the -Xconcurrentio flag, would that make a 
> significant difference? If so, is there any way to improve performance 
> on the windows platform?


--
Bela Ban
http://www.jgroups.org
Cell: (408) 316-4459