axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert J Walker" <>
Subject Re: Performance
Date Wed, 18 Jul 2001 16:57:02 GMT
I've been running some home made tests to determine throughput for a 
real application involving file attachments.
These tests were done under Windows 2000, single cpu, 9xxMHz, 512 MB 
memory. TomCat-Apache-SOAP.
A few observations:
The specific JVM can impact throughput by a significant amount, 17% on 
average for one installation scenario I did.
The fastest JVM depneds on the specific platform/version of the OS.
Surprisingly, when running on Windows 2000 Professional I got the best 
results with IBM (+17%);
when running on Windows 2000 Server I got the best results with Sun's 
JVM, again by a wide margin.
Bumping up process priority on Windows 2000 degrades throughput. This is 
counter intuitive. ?!
Reducing video resolution to 640x480 does boost performance.
Allocating more memory to the jvm also does boost performance.

At the end of the day I could reliably process 49 messages/second with 
attachments under Apache-SOAP.
i.e. between two workstations on a lan, in a well tuned environment.
By "process" I mean the server received and acknowledged that many 
messages in that time frame on average on twenty trials consiting of 100 
messages per trial.

One must be careful with Windows 2000 Professional. It does choke off 
your foreign socket connection requests when you reach some maximum 
magic number. Windows 2000 Server does not limit you on socket open 

Also there is a high degree of variability between the maximum and 
minimum observed throughputs.

I am looking formward to using Axis when attachments are eventually 

Robert Walker wrote:

> Interestingly, these are the performance numbers I'm currently getting with
> Sam's SoapRunner tests.  I'm using the latest CVS source of both xml-axis
> and xml-soap, freshly built today, running on top of JRun 3.1 on a dual-proc
> 800Mhz Kayak workstation.  I ran each engine once through tcptunnel to
> confirm that there were no errors, then got these results:
> messages per second      = 19.3
> microseconds per message = 51813
> messages per second      = 160.4
> microseconds per message = 6234
> These are averages after running the test 8 times on each codebase.
> I clock that as about an 8X performance improvement for this simple test.
> Sam had reported 2X... can other people run these tests as well just to see
> what the deal is?
> Glen Daniels
> Macromedia
> Engineering Manager
>                                 Building cool stuff for web developers

View raw message